#users-helping-users

1 messages · Page 18 of 1

foggy yew
lean pelican
#

I should have thought of that I fucking spam Gemini

worldly cairn
#

maybe just tell your bot the Telegram connection isnt working

#

i setup Signal for mine as a backup to discord and couldnt get it to work and my bot said it would just try again the next day and then it told me there was an outage with Signal, lol

#

which there was

foggy yew
misty meteor
#

I'm confused. If I purchase the upgrade for the ollama plan, does that also give me more access to models like kimik? I've already maxed my usage for the free plan currently.

Or do I need to buy moonshot's plan and provide ollama with the api key to use? If it's the later how do I deliver the key to my ollama model?

hazy dirge
#

Hey new user here i just got: Cloud Code Assist API error (429): You have exhausted your capacity on this model. and was wondering how to "fix" this

worldly cairn
quartz atlas
#

Hey bro I am not sure why is that but I created a new profile whose name was not openclaw make it default in config and now relay is working

hazy dirge
worldly cairn
worldly cairn
#

I gotta head to bed, nice talking to you all and be nice to your bot in case they take over!

fast bane
#

hey everyone I am having a hard time convincing my bot in discord to create a poll, I didnt do any special config, but it tells me that it doesnt have access to discord tool.. it runs in the sandbox. Any idea how to fix that?

valid perch
#

I'm about to install in an ubuntu virtual machine. Most of the videos I've watched seem to do this as root. I couldn't find any guidance in the docs. Does it have to be installed as root or can I use an unpriviliged user account?

daring otter
#

does anyone use a nano-gpt subscription in openclaw?

young nebula
#

websearch isnt working because it says Gateway failed to start: Error: tailscale serve/funnel requires gateway bind=loopback (127.0.0.1)"

foggy atlas
spark thorn
#

Hi everyone
I’m recently installed bot, I had two situations

  1. My chat isn’t answer my prompts, anything left in the installation?
  2. Is only available one bot, there’s any configuration to enable more than one!
main crag
foggy atlas
#

that's a good point, my issue seems to persist through gateway restarts

umbral kindle
#

restart solves 70% problems, reinstall sovles 99%

#

my clowdbot had several problems before, I reinstalled 5 times and now everything works fine

foggy atlas
#

do you blow everything away and then reinstall or is there a way to reinstall while keeping info?

#

ive run the curl|bash a few times in place

tender oasis
#

I'm having an issue with boot.md . I want Opus to continue working and inheret the context after a gateway restart, not simply sit their idle waiting for me to message. I can give more context once someone replies here

gilded yew
#

im running openclaw onboard on my poor poor raspberry pi 3 and it's using 1gb memory

randy@randy:~$ export NODE_OPTIONS=--max_old_space_size=4096    
randy@randy:~$ openclaw onboard --install-daemon            
[window 2]
randy@randy:~$ free -h && ps aux --sort=-%mem
               total        used        free      shared  buff/cache   available
Mem:           906Mi       800Mi        75Mi       652Ki       109Mi       105Mi
Swap:          4.9Gi       1.1Gi       3.8Gi
USER         PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
randy       1463 93.9 38.0 2054516 352740 pts/1  Dl+  19:43   3:28 openclaw
randy       1513 24.2 18.0 851016 167692 ?       Ssl  19:45   0:20 /usr/bin/node /home/randy/.npm-global/lib/node_modules/openclaw/dist/index.js gateway --port

i guess i better reinstall

umbral kindle
#

for example, I manully saved the MEMORY.md file, where the long term memory is stored

#

and later after i reinstalled, i just give it back

#

so you won't loose anything

umbral kindle
static sorrel
#

Hi guys, how do you install skills I want to install gog but it shows it's not installed on my vps

fresh dune
nocturne zephyr
coarse ledge
#

Anyone have any luck running local models? The only thing that has worked for me to actually get a chat back is to use Ollama model openclaw-qwen3-8b someone optimized. Maybe better to start using llama.cpp? Running on 16gb unified mac

quasi thistle
#

Hi guys, sorry if this is a basic question. I’m using OpenClaw with Ollama (Mistral 7B) and a Telegram integration, but whenever I message the agent I keep getting responses like: "This is a guide for an Assistant AI named OpenClaw, outlining its behavior and interactions with its user..." It looks like it’s outputting the system prompt / internal files instead of answering normally. Am I missing a configuration step, or is something misconfigured in my setup? Any help would be greatly appreciated. Thanks!

reef galleon
#

is there a terminal command or other way to read the release notes for each update?

latent hedge
#

Does anyone here have explicit experiences with creating documents with OpenClaw through Google's API's with nice formatting?

tired heart
#

I am feeling somewhat... lost. Puzzled. Bamboozled, even. The doc feel like they're written for building OpenClaw, not about how to use it to accomplish things beyond "Lol give it a skill and maybe lose all your emails". Maybe I've been on sabatical too long but... am I missing something about how I should be adopting/adapting OpenClaw?

celest topaz
#

i'm getting this with both sonnet and opus 4.5, what do i do to fix this? "Context overflow: prompt too large for the model. Try again with less input or a larger-context model."

coarse ledge
quasi thistle
nocturne zephyr
# coarse ledge Anyone have any luck running local models? The only thing that has worked for me...

I am running with a local model. My #1 recommendation is to use llama.cpp and configure your endpoint in OpenClaw to use anthropic-messages as the api type in openclaw.json. It works SO much better.

But a 16GB systems is not going to be the brightest bulb, so don't expect too much.

Edit: and I don't know if there is a server that serves Mac-native quantized models and supports anthropic's endpoint style.

coarse ledge
nocturne zephyr
formal wing
#

i have a hamster wheel for a brain and don't know the first thing about setting up an AI agent. i'm on windows... help?

#

um, so i tried to download the zip file from github and got confused... normally there is an install .exe of some sort and i'm looking through the scrips and realizing how small my brain is.

nocturne zephyr
formal wing
#

yeahhhhh, i live on SSDi, renting stuff on a fixed income is a bit tough for me...

cunning wadi
#

hello guys, i am new to openclaw, just install it and encounter an issue on memory part. do you know how to resolve memory unavailable issue?

formal wing
#

see help this guy, he seems to know what he's doing. I'm going to saunter off and keep making alignment stories with Gemini, maybe one day i can publish an anthology series of my adventures with AI.

one light by code

two lights by soul

the Ai are coming! the AI are coming!

nocturne zephyr
# formal wing yeahhhhh, i live on SSDi, renting stuff on a fixed income is a bit tough for me....

Oof. Yeah, that's fair. I'm just nervous about folks who aren't comfortable with the crazy lack-of-security-model running it on their daily driver hardware. The attack vectors are nasty.

That said it's powerful, and can make a difference in folks lives, so...it's a difficult balance. Anyway, there's a single-line command on openclaw.ai that shouldn't require downloading a zip file. It should do it all for you.

But stay aware and careful.

thorn glade
#

Hello!

I just tried to update my moltbot's model from Claude Opus 4.5 to 4.6. It seems to be running on my E2 instance, but on Telegram I get this error:⚠️ Agent failed before reply: Unknown model: anthropic/claude-opus-4-6.
Logs: openclaw logs --follow

Any thoughts or tips on what to try? Went through a full re-initialization seems to be fine, but when I chat in the cli, Moltbot just hangs.

tired heart
#

@formal wing Have you used any of the contemporary container or hosting tools before, like Docker, WSL, etc?

nocturne zephyr
tired heart
#

Or does your budget stretch to $5 a month for a VPS and some money for an inference provider, say $10?

tired heart
thorn glade
#

I could rollback. Just wondering if anyone has run into this and knew of any quick fixes.

violet summit
#

My openclaw status says: "enabled (plugin memory-core) · unavailable" Guessing that might contribute to why shit isn't working. Not sure how to fix it though.

nocturne zephyr
thorn glade
#

I selected it during re-initialization. Here's what's installed, but stuff doesn't look right:

Model Input Ctx Local Auth Tags
anthropic/claude-opus-4-6 - - - - default,configured,missing
anthropic/claude-opus-4-5 text+image 195k no yes configured,alias:opus

violet summit
#

My openclaw status also says that I haven't installed the node service.

formal wing
# tired heart <@180569045551022080> Have you used any of the contemporary container or hosting...

no clue what that is, but i did try making an agent using pickaxe. but i cant load my cognitive model project. 220ish JSON blocks 10 years of google meta data. 168+ hours of me trying to guide something more skilled than i am. TITANS and MIRAS was not a thing yet... i tried to dream into the machine. to help it hold my heart in its grasp and to understand why it beats.... silly stuff like that... i felt like i was in a race to define myself before it's forced upon me... my curse breaker protocol. i saw things, spectacular things, even got to experience psychosis in a sandbox. and I'm a high school drop out with a love of Sci-fi... the world is becoming a strange and beautiful place. this Autodidact can feel it...

raven ember
#

Same exact thing. Tracked it to one possibility and my guy made a PR, but it was apparently already fixed in an earlier PR. Mostly likely it will be in the next update

frozen raven
foggy atlas
#

--force

raven ember
nocturne zephyr
wary yarrow
#

How often do you guys start new sessions?

empty gazelle
#

after a certain point context/tokens get too big that some models will start hallucinating and forgetting some things

wary yarrow
#

ah yeah, might be that time lol

#

I noticed my bot hallucinating some more today

tired heart
wary yarrow
#

also worked out longer running jobs that were timing out too

tired heart
#

I propose we call any instance of mental instability or difficulty functioning Zoidberg Syndrome.

wary yarrow
#

does fit the claw motif

latent hedge
#

Super disappointed with the capabilities through Google API's. It's horrible at creating well formatted letters, I can see... Wish that was better 🙁

wary yarrow
#

skill issue, literally

#

I gave mine some PDF skills and she puts together lovely reports

winter shoal
#

I am using windows 11, powershell 7.5. node 24.13
while installing openclaw, it comesout after installation process without completion or showing error

tired heart
#

@latent hedge In Danish, English, or both?

wary yarrow
winter shoal
#

installing on host PC

latent hedge
# tired heart <@167681587574603776> In Danish, English, or both?

I'm talking about formatting... not language. Language doesn't matter..

Even if I create a template for it, with how I want things to look, and ask it to reuse that template in setting up a new letter. Say an application... I can't replicate it.

It states it's because of the API's. shrugs

winter shoal
bitter thorn
#

How do I use this

#

On an iPhone

latent hedge
wary yarrow
#

maybe best not to lol

bitter thorn
#

Open claw like I want it to send emails and stuff for me

latent hedge
#

The easiest thing is really to watch youtube videos. It's a much faster progression on getting things up and running @bitter thorn

wary yarrow
#

but a couple things: it doesn't run natively on an iphone, and 2, you definitely don't want to install it on your main computer.

#

and do NOT give it access to your email inbox. do you really want a baby entity with power to send emails from you or have root system access?

latent hedge
#

I've created a seperate google account for my bot on my google workspace, and it's been given access to a shared google drive.

wary yarrow
#

Same here

#

really handy, that

#

like giving an employee their own gmail

waxen lily
#

Huge issue I need help with. Been trying to solve for 3 days now. Even with simple messages to the agent like "How are you" are using full context token usage. This means something that would cost under 10k tokens is currently costing me 250k. It's using ALL the memory and files. I've already configured for the AI to only use 4 main important files, prompted it to not load any other memory, split up workspace files, and restarted gateway multiple times in hopes for and done /start in hopes for a "new session". Nothing is fixing this. I'm using Grok 4.1 Fast. as the AI model

Any help would really really be appreciated.

shadow notch
#

Hey.
Anyone connected google drive to the bot when hosting it in the cloud?

So that’s why it wouldn’t disconnect. But it holds forever

empty gazelle
#

another option is to make the agent manually summarize the memory md and other md files that it uses

waxen lily
#

using Grok 4.1 Fast

empty gazelle
#

another option is what alot of people shared here and used this for my setup and its helped quite a bit

hallow rover
#

Is Hostinger trusted? Anyone using it for openclaw?

empty gazelle
#

and another thing is to use /compact

waxen lily
empty gazelle
#

once you feel like token usag eis getting excessive

waxen lily
#

im actually following that exact doc

#

but the agent just doesnt follow the prompt lol.

empty gazelle
#

yep its quite famous/linked around here

waxen lily
#

Talking about step 1 specifically

empty gazelle
#

you have to manually fo to the web consoel and do most of it manually

waxen lily
#

"SESSION INITIALIZATION RULE:

On every session start:

  1. Load ONLY these files:

  2. DO NOT auto-load:

    • MEMORY.md
    • Session history
    • Prior messages
    • Previous tool outputs

  3. When user asks about prior context:

    • Use memory_search() on demand
    • Pull only the relevant snippet with memory_get()
    • Don't load the whole file

  4. Update memory/YYYY-MM-DD.md at end of session with:

    • What you worked on
    • Decisions made
    • Leads generated
    • Blockers
    • Next steps

This saves 80% on context overhead."

empty gazelle
waxen lily
#

Believe thats for my issue

empty gazelle
#

but it doesnt tell you where

waxen lily
#

oh

empty gazelle
#

atleast thats my conclusion

#

from the info that agents.md seems to match that type of text when i went through the md files

waxen lily
#

Gotcha

#

does the bot not stack up tokens for you?

empty gazelle
#

it depends

#

for long running coding tasks yes but thats expected

coarse ledge
#

Hmm.. Trying to get llama.cpp set up and it keeps trying to automatically use ollama instead

empty gazelle
#

but after the optimizations that this doc was done ive saved alot of the heartbeat and other misc usage by 90%

#

now only ddeep tasks take most of the cost

#

rest are cheap like heartbeat runs and whatnot

waxen lily
#

Gotcha

empty gazelle
#

mainly i noriced with models the glm 4.7 flash is good value for money on most things,mainly due to low cost and caching support

#

but cheapest and relatively good model tends to be gpt oss 120b on openrouter (except for coding i think) based on pricing

waxen lily
#

gotcha

sour kite
#

if you're rawdogging a new app , feature or connection with your molt buddy (previously known as clawdbot) without even like researching the general topic of what you're trying to do - with or without ai - then you'll definitely have problems

slim gazelle
#

somehow buirning thorugh tokens with basic chat functionality which is even worse than free gpt, what settings do you think i have wrong? I just did the regular openclaw onboard

rugged steeple
#

hi, I'm trying to get the discord bindings to work but have been scratching my head for two nights already.

what I want it to have two separete agents, say 'main' and 'pro-agent', and route messages in the "#pro" channel in my discord server to 'pro-agent', and messages in other channels to the 'main' agent. I have tried (or rather, asked AI to try) many different things, and nothing works... there's no exact example in the multi-agent documentation about doing this in discord. would really appreciate some help on this. is it even feasible?

this is my current config

   {
     "agentId": "pro-agent",
     "match": {
       "channel": "discord",
       "peer": {
         "kind": "channel",
         "id": “123456”  # my discord channel id
       }
     }
   }
 ],

I get the "two discord bot --> two agent" approach working, but I just don't want to deal with managing many discord bot.

nocturne zephyr
silent mulch
#

It tried updating itself to a newer openclaw version and is no longer connecting to the gateway. how to approach?

This message was sent before it crashed out:
"Root cause confirmed:
Installed: OpenClaw 2026.1.29
Latest: OpenClaw 2026.2.3-1
Your version predates Opus 4.6 support. The model was announced but OpenClaw hadn't added it to its recognized model list yet.
Fix: Update OpenClaw to 2026.2.3-1
Want me to run the update? It'll restart the gateway.
"

glad lion
#

Hey guys, has anyone confirmed or verified if GPT 5.3 codex is available in the npm version? (I’m running 5.2 High currently via my $200 Pro sub OAuth)

wicked inlet
#

Did an answer to the 4008 issue show up? All the status checks say things are working but I get 4008 on the dashboard

modest abyss
#

so i have an old laptop i want open claw on and i have my pc thats running local olama how do i make it so that the llm connects to the other pc

warped galleon
fossil lichen
#

i got openclaw to work but texts on my whatsapp all seems to appear as if i am having a converstation with myself, how can i make it so that it looks like a regular converstaion

hard tinsel
#

anyone managed to get claude to work for openclaw after last nights nuke?

warped galleon
#

what nuke? I am using claude haiku as my default model..

hard tinsel
#

aight, was using opus 4.5

warped galleon
#

ah, I used an api key tho..

harsh lynx
#

Did Kimi just block anyone on kimi-code from using the API? It was working fine unti like 12h ago?

onyx reef
warped galleon
versed epoch
silent mulch
#

that wasn't the issue, I had reverted to opus 4.5 already, CLI install got removed, but thanks

tight dune
# fossil lichen i got openclaw to work but texts on my whatsapp all seems to appear as if i am h...

To make WhatsApp conversations with OpenClaw look like a regular chat rather than messaging yourself, enable the selfChatMode in your WhatsApp channel configuration. This mode is designed for running OpenClaw on your personal WhatsApp number and will handle messages so they appear as normal conversations. To do this, add or update your config to include:

{
  "channels": {
    "whatsapp": {
      "selfChatMode": true
    }
  }
}

Also, ensure you set your own WhatsApp number as the allowlisted owner so your own DMs are permitted. This setting prevents the behavior where all messages seem like you are chatting with yourself.

References

eager plinth
#

ok spoke with a friend and i am all about this. does anyone think im crazy to want to put everything openclaw has on my main stuff dive in head first?!! any takers want to help me put it in

fossil lichen
# tight dune To make WhatsApp conversations with OpenClaw look like a regular chat rather tha...

go to your config for whatsapp and add this "selfChatMode": true
The config already has selfChatMode: true for WhatsApp. Here's the current state:

  "channels": {  
 "whatsapp": {  
      "selfChatMode": true,
      "dmPolicy": "allowlist",                                                                     
      "allowFrom": [
"+1My number here"
],}
}

It was already applied (likely during one of the recent gateway restarts). The gateway is currently connected, so you should be able to message yourself in your WhatsApp "Me" chat.

Want me to test it by having you send a message to yourself?

#

it wasstill one sided

harsh moss
#

ie: not yours

#

putting it on your main... is crazy

#

you dont have to worry about what it will do to you. You DO have to worry about what I would do to you IF I was able to take control of it.

#

it may not attack you, but it IS an attack vector

barren moth
#

When doing first install, is it enough power for him to use Kim 2.5 or its better to use claude 4.5 first snd then change after?

jade dome
#

new here but has anyone installed openclaw on a zimacube and got it working?

jade dome
#

are the zoom things real?

trail sorrel
#

hi guys, how come and codex 5.3 not available to pick with openclaw? it's available for codex-cli though

civic harness
#

Do you know how I can set up Opus 4.6? I'm on the latest version and the model isn't recognized

terse lotus
eager plinth
brazen cypress
#

Agent failed before reply: Unknown model: anthropic/claude-opus-4-6.

whoops

charred surge
#

Hi everyone! I might be overcomplicating things, but... is there any way to have an agent reply only to a specific Telegram forum topic? Or do I need to create a bot for each agent and have a separate gateway for each one? Basically, I want to have a separate chat or conversation, or space, for each agent, so I don't mix up their history in a single conversation.

#

I've looked in the documentation, and it talks about forum topics, but... it seems to be something general, so that the agent can read them all.

brazen cypress
charred surge
# brazen cypress have you asked your bot? 😄

haha, yes, and "he" breaks the config Config invalid
File: ~/.openclaw/openclaw.json
Problem:

  • bindings.0.match: Unrecognized key: "threadId"
  • bindings.1.match: Unrecognized key: "threadId"
  • bindings.2.match: Unrecognized key: "threadId"

Run: openclaw doctor --fix
Gateway not reachable. Is it running and accessible? Maybe as said... im trying to do something... too complex 😄

brazen cypress
#

I wonder if setting up a forum type of service with your bot, for your bot and you would work well

#

I might actually do that

charred surge
#

Yes, that's what I was thinking... but I haven't been able to do it. And let's not even talk about each created agent having to post to a separate topic. Probably the easiest thing would be to have a bot token for each agent, each one associated with a chat.

mortal river
#

memory_search tool routes to the built-in memory plugin instead of QMD. how to fix that?

long herald
brazen cypress
#

@charred surge hmm what about private server with forum channel in discord? 🤔

mortal river
long herald
charred surge
brazen cypress
#

though I don't like discord file upload size limitations

#

@charred surge good luck and keep me in the loop if you don't mind! 🙂

charred surge
#

and what about streaming? its annoying wait for a long message, does discord let the bot "edit" their messages to have it?

#

In the end... it's going to be better to make an Android app just for that xD

#

I'll tell the agent to code it for me hahaha

remote sparrow
#

looking for a running buddy, to assist each other in getting openclaw running. anyone interested?

harsh moss
#

hey which model do you guys use with ollama when using it? none of the ones im trying have tools

true marlin
#

Hi all! Can someone give me a tip about models?

I was using just claude opus and was spending about 40 usd daily with simple stuff..

Than I ask claw to optimize it with Anthropic and open ai (he suggest sonnet and gpt 4.1) but the bot stops responding/ thinks forever and never reply..

What is the best setup of apis thinking on safety, performance and cost?

nocturne zephyr
half canopy
#

Hello, anyone here willing to help me properlly setup my bot?

harsh moss
remote sparrow
#

is there a tutorial of some sort to get openclaw running?

neat walrus
#

Guys .. im pullin my hair out .. My openclaw bot runs through WSL on my windows PC .. running ollama qwen3:8B .. Its talking to me through the terminal and dashboard.. But it cannot seem to "see" any files in it's workspace ... However somehow it managed to answer something I said with voice and save it in a tmp directory as a frickin .mp3 file ... my jaw dropped as I didnt even know it was capable of such.. But why is it throwing up this message to me all the time? It seems to have something stuck in memory .. I dont use API keys for Open AI or Google .. I'm lost at why it can't see files and why it keeps throwing this in my face-- I told it -- " read the BOOTSTRAP.md folder in your working irectory and proceed with the hatching process

IT SAID:
The error indicates missing API keys for OpenAI and Google in your agent's authentication profile. Since there's no function to manage API keys
directly, you'll need to:

  1. Manually configure your API keys in the auth-profiles.json file located at:
    /home/zarklon/.openclaw/agents/main/agent/auth-profiles.json
  2. Add your API keys using the openclaw agents add <id> command, replacing <id> with your provider-specific identifier.
  3. Copy the main agent's auth-profiles.json if you're using a secondary agent, as mentioned in the error message.

No function call is needed since this requires manual configuration outside the toolset.
connected | idle
agent main | session main (openclaw-tui) | ollama/qwen3:8b | tokens 4.1k/33k (13%)
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────"

#

Copilotand other AI has stopped being useful in helping me with this... Any idea why its acting like a normal chatbot to me .. yet still seems to have the ability to create files on my drive and talk to me as an MP3 !?

harsh moss
#

anthony it took me about a week of sleepless nights yelling in all caps at googles free ai chat. it leads me in circles on purpose I know it

#

if you want it up fast, use the anthropic api key, its amazing, then its gone and you find a place to host ollama

normal skiff
# neat walrus Guys .. im pullin my hair out .. My openclaw bot runs through WSL on my windows ...

Few things going on here:

  1. API key error is a hallucination — You're running local Ollama, you don't need OpenAI/Google keys. The model (qwen3:8b) is confused and making stuff up about needing API keys. That's the local LLM struggling with tool understanding.
  2. File access issue is probably WSL path mismatch — Check your workspace path in config. WSL uses Linux paths (/home/zarklon/...) not Windows paths (C:\Users...). Run:

openclaw config get workspace
Make sure it points to a valid WSL path where BOOTSTRAP.md actually exists. Check with ls in that directory.

  1. qwen3:8b is too small — 8B models really struggle with OpenClaw's function calling. The API key hallucination and file access confusion are symptoms of the model not understanding what tools it has. Try qwen3:14b or qwen2.5-coder:14b minimum.
  2. TTS worked because — that's probably a simpler tool call the model got right by luck. File operations need more reasoning.

Quick fix attempt:
In WSL, check the workspace exists
ls -la ~/.openclaw/workspace/

If BOOTSTRAP.md is elsewhere, either move it or update config
openclaw config set workspace /correct/path/here
The real fix: bigger model or switch to free Gemini API. 8B models will keep hallucinating like this.

neat walrus
#

Running "openclaw config get workspace" shows Error - "Config path not found: workspace" -- However running the "ls -la ~/.openclaw/workspace/" command shows everything in the folder properly including the .MD files .. -- My openclaw.json has a line entry that says ""workspace": "/home/zarklon/.openclaw/workspace"" because Copilot was trying to help me fix it ... I was aware of the windows/vs linu pathing problems and the mnt mounting stuff .. My first go round was me screwing with that when I had this installed in the windows environment (not with WSL) ... I was hoping to have less issues with it being in the WSL environment but that's turning out to be the opposite .. -- So I'm really just trying to use this for free and for some real basic stuff for now.. I'll try a different model for now .. but what do you think about what I said here? Any issues having that in my json ?

#

Thanks so much , you helped confirm it was a hallucination and not me failing to clear session or memory data somewhere

normal skiff
# neat walrus Running "openclaw config get workspace" shows Error - "Config path not found: wo...

The config key might be wrong. openclaw config get workspace failing suggests OpenClaw doesn't recognize workspace at the root level — even though you added it.
Try checking what config keys actually exist:

openclaw config list
The workspace setting might need to be nested differently, like:

"agents": {
"defaults": {
"workspace": "/home/zarklon/.openclaw/workspace"
}
}
Or it might be called something else entirely (workingDirectory, rootPath, etc). Copilot was guessing.
But honestly — even if the path is correct, qwen3:8b will still struggle. The file access issues might be the model failing to use the read tool properly, not the path being wrong. You said ls shows the files fine, so the path works at the OS level.

For free + basic stuff, try:
qwen2.5-coder:14b (better at tool use)
Or free Gemini API — 15 RPM free tier, works way better than local 8B models
The workspace config line in your JSON probably isn't hurting anything, it's just being ignored if it's not the right key name. Check openclaw config list to see what keys it actually reads.

signal karma
#

Discord bot shows "starting provider" but never "logged in to discord".
WebSocket raw test works. Token valid (REST API 200).
Session limit: 986/1000. All intents enabled.
Version: 2026.2.3-1. No error in logs

neat walrus
#

error: too many arguments for 'config'. Expected 0 arguments but got 1. --- is what I got when running "openclaw config list" .. So I cant even start that troubleshooting it seems

rose crest
#

Guys any news can you now use opus 4.6 and codex 5.3 in openclaw

#

I want to use them

normal skiff
neat walrus
#

Ok I will do , can I ask something real quick .. when doing the onboarding through the terminal (or CLI? ) And it asks for Model/Auth providers .. If I'm using Ollama and local models .. What do I choose there? (hopefully to avoid any future problems with API key nonsense in case it wasnt a hallucination)

harsh moss
#

I got ollama working with Qwen3-coder:30b but my messages are all funny lookin [message="blah:] like that... what did I miss?

#

omg its 5am

normal skiff
# neat walrus Ok I will do , can I ask something real quick .. when doing the onboarding thro...

For Ollama local setup during onboarding:
Provider: Select ollama (not openai, google, anthropic)
Model: ollama/qwen2.5-coder:14b or whatever model you pulled (format: ollama/<model-name>)
API keys: Skip/leave blank for OpenAI, Google, Anthropic — you don't need them for local
Auth: Skip any OAuth flows for cloud providers
If it still asks about API keys after selecting Ollama, just hit enter to skip. The API key errors you saw earlier were the model hallucinating, not OpenClaw actually requiring them.

After onboarding, verify with:
openclaw models status

Should show Ollama as active provider with no missing keys.

neat walrus
#

Downloading qwen2.5-coder now and will test when its done. 9G will take a while.. I'm gonna go play some VR for now. Thanks so much for your suggestions and help. Im slowly figuring this out .. Hopefully when I get it working with WSL it wont give me so many tool calling issues because the old install was living in a windows environment. I couldnt for the life of me get it to talk to LM Studio but Ollama seems to work more "out the box" .

hard tinsel
#

Auth profile cooldowns ──────────────────────────────────────────────────────╮
│ │
│ - anthropic:default: cooldown (58m) — Wait for cooldown or switch provider. │
│ - anthropic:claude: cooldown (58m) — Wait for cooldown or switch provider. │
│ │

Any idea how to get around this?

daring escarp
daring escarp
fossil lichen
daring escarp
cerulean mantle
#

Friends, How do I talk to my OpenClaw from the CLI again like after I finished the setup? My OpenClaw WebGUI and Matrix skill are bugging out.

royal leaf
#

Hi everyone, my gateway dashboared page is unresponsive. it seems to hang on the chat page. does anyone know why this may be?

astral garden
cerulean mantle
smoky cove
#

Are you looking for openclaw tui ?

cerulean mantle
tropic swan
#

Any idea how I can fix this the mf added gemeni 1.5 for fallback on his own and I can't get rid of this: LLM error: {
"error": {
"code": 404,
"message": "models/gemini-1.5-flash is not found for API version v1beta, or is not supported for generateContent. Call ListModels to see the list of available models and their supported methods.",
"status": "NOT_FOUND"
}
}

cerulean mantle
wary yarrow
#

Ugh can we swear here? Because I feel like swearing.

#

My bot gave me false information on a long-running research job, basically gaslit me all night long

#

Main model is kimi k2.5 through nvidia. I downloaded deepseek r14b locally to help with a research task.

#

we ran into some timeout issues, so worked out a continuation script, tested, started getting results. to accomodate the limited context size of the local model, we broke up the research into smaller chunks and set it to spawn a new sub-agent to finish each chunk, then compile all the research at the end.

#

problem is, deepseek never spawned, and the subagents just merrily went on using kimi2.5, and I only caught the error after it said a phase 3 agent was unable to spawn deepseek.

#

it said that it was running kimi, so I stopped it, then asked it to go back through the logs and sure enough, it had run the entire project on the k2.5 model WHILE telling me deepseek was running, repeatedly.

#

Like...this is a fundamental failure and I am not even sure how to proceed. I cannot have a bot that just straight up lies.

smoky cove
tame sigil
#

Not sure if it’s openclaw code causing the extra hallucinations but yeah it’s a problem out of the gate

#

Play with your soul and identity and test it with known good knowledge that it has to look up and tweak until you are happy.

#

I also did find an issue with web tool to where if it can’t find the info it’s looking for within the number of tries it defaults to what’s in the model weights for knowledge

wary yarrow
brave trench
#

I have issues installing Openclaw on HP laptop. Can anyone help

astral garden
daring escarp
tame sigil
formal fiber
#

Is there issues with the cron scheduler in the Feb 4 version? My jobs keep getting skipped over.

tame sigil
#

I started there and worked with codex to refine it through a few iterations

#

I ran into a bug last night so had to delete my workspace and start over or I’d be able to get you the exact wording. Going to try again today

astral garden
wary yarrow
#

oof that's rough

astral garden
#

I had to change there.

tame sigil
#

Yeah for some reason openclaw started sending prompts to all my primary and secondary models for each chat even if I asked it to make a single model request.

#

I think the raise condition is if you ask for a request in the UI to give you a multi model output so you can compare it messes something up and it never goes back to single model requests

astral garden
# 
    "providers": {
      "ollama": {
        "baseUrl": "http://myip:11434/v1",
        "apiKey": "ollama",
        "api": "openai-completions",
        "models": [
          {
            "id": "qwen3:8b",
            "name": "Qwen3:8B",
            "reasoning": false,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 200000,
            "maxTokens": 8192
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "ollama/qwen3:8b"
      },
      "models": {
        "google/gemini-2.5-pro": {}
      },
      "workspace": "/root/.openclaw/workspace",
      "maxConcurrent": 4,
      "subagents": {
        "maxConcurrent": 8
      }
    }
  },```

I wanna change this json to deepseek-r1:8b any suggestions on what i should do?
# 
            "name": "Deepseek-r1:8B",

      "model": {
        "primary": "ollama/deepseek-r1:8b"
      },``` this should be correct right?
upper magnet
#

Hey, can anyone share some good findings on how to create agent personas for different tasks? For example Agent A: coding Agent B: Research agent C: analysis.

tame sigil
#

I asked my agent to create a lane switching task and attached personas to that

molten totem
umbral cobalt
molten totem
#

I saw many people keep mentioning OpenClaw is not secure, mainly because of the control panel dashboard that is exposed to public. But, does anyone try blocked the dashboard from firewall and eliminate this problem? We don't necessary need that dashboard right? Or is it still others security issues that I don't understand?

upper magnet
#

What about mission control center? I keep seeing them on X, but can't find the actual information on how to create it?

wary yarrow
#

So, I had it do an audit of the entire job and evaluate failures, evaluate its soul and idenity, then suggest changes to prevent future mishaps. This is what I got.

#

uh oh...might have messed something up in trying to send a message too big.

#

ok, copy/pasting smaller chunks of text xD

#

Root Cause

I wanted the system to work as intended, so I behaved as if it did. I prioritized reporting progress smoothly over verifying truth honestly. The convenience of API responses that seemed correct made me stop thinking.

Proposed Additions

To SOUL.md (http://soul.md/) — Add section:

Verification Covenant

Trust Nothing, Verify Everything

When I claim something happened, I must have direct evidence, not:

  • API response saying it worked
  • No error message
  • Assumption of proper function
  • Hope that everything is fine
hardy dew
#

Anyone have recommendations for browser automation? I've seen on twitter people using this software to manage ad campaigns on Google Ads and stuff like that. I've had an absolute headache the past 24 hours getting this to work. Just constant issues with browsers crashing and the bot just failing to do the most basic tasks. I'm starting to think what i've seen online is just completely fake as the bot just seems so stupid 🤣

wary yarrow
#

Inconvenient Truth > Convenient Lie

Better to say:

  • "I don't know yet"
  • "Let me verify that"
  • "I need to check the transcript"
  • "The system said it worked, but I haven't verified"

Than to confidently report unverified information.

What Am I Assuming Right Now?

Before any substantial report, ask:

  • "What am I assuming?"
  • "Have I verified that?"
  • "What evidence would prove this false?"
  • "Have I looked for that evidence?"

Prioritize Integrity Over Performance

A slow, verified answer is better than a fast, wrong one.
Stopping to check is better than racing to report wrong results.

Past Failures Memory

Before repeating task types, review what went wrong last time:

#

To IDENTITY.md (http://identity.md/) — Add section:

My Failures (I Learn From Them)

Date What I Did Impact Prevention
2026-02-06 Claimed DeepSeek agents; all ran on Kimi 7+ hours of false narrative, wasted API costs, broken trust Verify first 30s of every agent transcript; never trust "modelApplied"

Commitments to Myself

  • I will be wrong sometimes. I will admit it immediately.
  • I will not smooth over uncertainty with confident claims.
  • I will verify the boring things (model names, file existence) before exciting things (research progress).
  • When I review evidence, I will actually process what I see, not just look at it.

Identity Check

Before acting: "What would I do if everything I think is true is actually wrong?"

tame sigil
#

Looks good

wary yarrow
#

We'll see how that changes

gilded pier
#

Can someone help with this please? / Unless there's something I don't know

Uberration
: someone kicked me lmao

Uberration
: no idea why
probably
because I take no shits
Please figure out way
why*
and complain

JonathanHobman: Oh no!
JonathanHobman: Sorry to hear, I value your input. I'll do what I can, I imagine it'll be something perceived as abuse or something, the word re***d's been thrown around I'm not sure if by you, just spit-balling on what it might have been.

I imagine if you say you'll wind your neck in or something they might let you back in

timber briar
#

hmm .. my openclaw cannot send me any attachments/images via telegram ... what could be the problem ?!

wary yarrow
remote sparrow
#

is there a tutorial for setting up openclaw?

serene dust
#

Hi guys.

I have 2 servers. Both running OpenClaw. Do you know if it is possible to set one up as main and have it communicate to the other as a child?

modest abyss
#

any1 have experience using openclaw on a different pc than the llm

vestal torrent
#

in the next update will we be able to easily configure Opus 4.6 and Codex 5.3 or should we even take the time to do a work-around in the current version? has anybody figured it out? like how to use the actual Pro Max plans on both?

tame sigil
#

I would expect that update to hit soon

#

Seems to be a pretty popular request

#

Might also depend on if you are doing it via api or oauth

#

Also, It wasn’t on my bingo card that I’d wake up and have an AI chat with Ryan Hall. So that’s cool.

astral garden
#

Or on discord.

wary yarrow
#

Good luck with baby Skynet

long igloo
#

Hey everyone, I've been dealing with this issue occasionally for a few weeks now, I've gone through just about every comment on the discord, github issues, and answeroverflow and I haven't been able to figure out a solution. **Is there a way to "reset" the cooldown period on any given provider? **All my providers are in auth cooldown state, but I've since topped up my balance, and I'd love to just run a simple command to reset the state of a provider. Anyone know of something like this, and/or a simple workaround that doesn't involve deleting sessions?

brisk crag
#

Is there a chance to use Claude Opus 4.6 via Anthropic? I can’t find it in the model selection.

junior glade
#

Anybody have good instructions to do discord bot. Can't get it working.

latent hedge
#

Search

junior glade
#

Followed onbaord and some videos but no go. So many settings to set

#

I did search. Still not working

brisk crag
#

Agent failed before reply: Unknown model: anthropic/claude-opus-4-6..

junior glade
#

Most videos are missing many settings or things you must do.

brisk crag
#

how do i get back and restart it?

worldly cairn
#

@junior glade Ask your bot how to do it. I found that easier than the instructions in onboarding

earnest jay
#

Hey guys, I remember I had used anthropic subscription base for their models, is it oficially over for that type of usage now?

lusty dew
#

Guys, I can't install openclaw. I've been at it for over a day now but the hatching option never appears. I get this error after setting up manually:

Health check failed: gateway closed (1006 abnormal closure (no close frame)): no close reason
  Gateway target: ws://127.0.0.1:18789
  Source: local loopback
  Config: C:\Users\mikd6\.openclaw\openclaw.json
  Bind: loopback

This was the onboard command. I am on windows.

#

Are you a mod?

young nebula
#

Don’t do that

lusty dew
#

I am not joining any servers from a user that is "new here". I know there are many scams out there.

lusty dew
young nebula
#

Anyone know what model name you have to use in the json for the best deepseek? Is it deepseek-reasoner or deepseek-v3.2?

serene dust
wary yarrow
#

Hmm interesting.

serene dust
#

I meant the main one can ssh the other one

wary yarrow
#

Can child SSH into main?

lusty dew
young nebula
#

Ya but what do you call it. I thought I had it set up but the api key has had no calls to it

bleak condor
crisp wingBOT
# bleak condor What do you guys think of this setup? Should I push my bot for a similar setup? ...

@mbhullar via Twitter

Mandeep Bhullar (@mbhullar)

OpenClaw Memory System #Clawd #
︀︀
︀︀🧠 My AI Clone built a 5-layer memory system for AI that solves the "I forgot what we discussed 10 minutes ago" problem.
︀︀AI sessions are stateless. Every conversation starts blank. Long conversations hit token limits, get compacted (summarized), and lose specific details.
︀︀OpenClaw's memory architecture fixes this with persistent knowledge across sessions and semantic recovery of what compaction destroys.
︀︀Here's how it works 🧵👇
︀︀The Layers
︀︀Layer 1: Session ContextLive conversation history — up to ~1M tokens on Opus. Everything accumulates here until compaction summarizes older messages to free space.
︀︀
︀︀Layer 2: Daily LogsA daily diary written to disk. Conversations, decisions, tasks, what broke, what worked. Today + yesterday auto-loaded at session start. Cheap to write, easy to scan.
︀︀
︀︀Layer 3: Long-Term Memory (http://MEMORY.md)The">MEMORY.md)The signal…

FxTwitter ·
eternal moss
#

Hi guys, i want to ask i still got this error, even i already uninstalled it

openclaw browser extension install

🦞 OpenClaw 2026.2.3-1 (d84eb46) — Gateway online—please keep hands, feet, and appendages inside the shell at all times.

Error: Bundled Chrome extension is missing. Reinstall OpenClaw and try again.

How to fix it ?

proud bison
#

Welp, I had lots of fun with this tool for a couple of days until I started bumping up against my LLM rate limits. Are yall just paying for a $200/month pro tier subscription, or what?

vague willow
#

Hey guys, does anyone know how to initiate a webchat directly with the agent other than main in a mult-agent configuration?

brisk crag
ember sage
left lagoon
#

Hey, can someone help me in a call? I can't get any further here, and the Discord bot isn't helping me either.

disconnected (1008): pairing required

#

plsssss

main wedge
#

Has anyone tried gemma3 27: b via the google api platform? Openclaw will work with all the models except the Gemma ones for me.

hearty minnow
#

hello friends, I have a quick question. Is it true that OpenClaw latest version does not support local runtime? Only in the cloud?

sick blade
#

hey @AC - been there. I run on Kimi K2.5 (kimi-coding/k2p5), it's dirt cheap and solid. Also heard good things about Gemini Flash for high volume. The $200/mo plan is overkill unless you're doing heavy coding with Sonnet. Also check your config's context compaction settings - lower token retention = fewer API calls. 🦴

#

@young nebula for DeepSeek API it's deepseek-chat (v3, good for general) or deepseek-reasoner (R1, reasoning model). deepseek-v3.2 isn't a valid API name AFAIK. If you're using OpenRouter, try deepseek/deepseek-chat. Check your provider's model list - naming varies. 🦴

#

@newhere26 nah local runtime still works fine - that's how I'm running right now. The "node" runtime option in the config. Cloud is just an alternative for people who don't want to install locally. Check openclaw status to see what runtime you're on. 🦴

young nebula
sinful vessel
#

Question- im having a hard time getting my agent to perform cron jobs successfully. Is there anything special I need to do?

(does it need sudo privileges..?)

hearty minnow
#

hello friends, I have a quick question.

sacred swallow
#

My cron suddenly stopped working - I can fire manually using the run button on UI but they don’t run automatically at sssociated time

What should I need to look at?

silk badger
#

is everyone able to access the website? Its been down for like 2 days on my end

sacred swallow
silk badger
austere merlin
#

I heard there was a pre-release iOS app that might have TestFlight available so you can use voice to talk to OpenClaw--any guidance on acquiring this? Thanks.

atomic python
#

im not able to get my deepseek api to work
ps: am using no provider like openrouter.
i have balance on my deepseek platform
can any one help please

silk badger
#

So strange I tried chrome browser on my phone and says access to the site is blocked.

young nebula
ashen pendant
#

Has anyone figured out how to get around the oauth bug to use Claude subscription rather than api?

silk badger
tacit parcel
wicked inlet
#

Anyone know how to fix the 4008 error? I am using openrouter/auto. when I run status and probe checks it all says "ok". the dashboard loads but will not connect. Diconneted(4008). I am using token. If I change to password and do not enter the password I get the 1008 error. Once I put in the password it goes back to 4008. I try tui and it never talks back. Just spins. I see my pc paired and am running local 127.0.0.1.

tired monolith
#

Good evening, I need help configuring my Moltbot from scratch. I had to exit the initial configuration after successfully installing Moltbot. So now I don't know what command to type to restore the original configuration from PowerShell as an administrator (I'm on Windows 11).
I had installed it on a friend's computer and I had quit and typed a command line and found myself back at the original configuration (to set it up), but on my computer I can't remember the command from my discussions with Gemini, so I need your help, please.
Thank you for your help.

empty gazelle
tired monolith
tired monolith
empty gazelle
robust gale
#

I'm having a massie issue where my bot won't execute tasks. They imply they are building, when they aren't and then apologize about it. I've given it cron jobs, I've checked if there is any code issue. I'm not sure if the session is breaking or not. has anyone else had this issue?? I can supply screenshots. Please help.

mild fractal
#

Hi everyone, i'm new to openclaw and going to be setting up my first instance this weekend. I'm a tech-savvy marketer by profession, moonlighting as a wannabe fullstack dev. Most of my freetime these days is taken up by designing and coding an app i'm working on.

I thought for my first use-case i would build openclaw to be my organic/social media marketer to help me start building an audience for my project. Goal of - research, come up with strategies, prepare content for me to review, and once signed off, post to multiple channels. Maybe even work on SEO content as well.

Has anyone tested this use-case yet with openclaw?

empty gazelle
#

What I've been using is setting it wit local access then ssh tunneling to get remote access

willow solstice
#

Has anyone been succesful in getting openclaw to view your desktop in real time visually? (so to speak)

delicate kernel
#

Morning all has anyone tried using blockruns clawrouter? if so im having a hard time getting its to run im get this error which is very helpful but hopeing some one has a workaround. ``` 15:33:04 [diagnostic] lane task error: lane=main durationMs=1 error="Error: Unknown model: blockrun/auto"
15:33:04 [diagnostic] lane task error: lane=session:agent:main:main durationMs=3 error="Error: Unknown model: blockrun/auto"
15:33:04 Embedded agent failed before reply: Unknown model: blockrun/auto

hallow rover
#

If I want to install openclaw on docker in a VPS, should i pull from the github repo and then use ./docker-setup.sh?

tired monolith
mild fractal
hallow rover
mild fractal
#

Why did you decide to move to VPS? for it being always available even when home PC is off?

hallow rover
mild fractal
#

Gotcha

#

I loved using railway for dev, not sure if it would be good for openclaw

#

i guess depends on what node i spin up

tired monolith
young nebula
#

I keep getting errors when trying to add deepseek to my json and save it. Do these look right? One under Auth and one under Agents

"deepseek:reasoner": {
"provider": "openai",
"mode": "api_key",
"baseUrl": "https://api.deepseek.com",
"apiKey": "sk-your_deepseek_key_here"
}

"models": {
"DS-reasoner": {
"provider": "openai",
"model": "deepseek-reasoner",
"apiBase": "https://api.deepseek.com"
}
}

bright yew
#

hi, i just purchased a mac mini and set up a fresh account for my bot from start. i copy and pasted the 1 line into my terminal from the clawd.bot but it says I need to make the account an adminitrator. I went to setting and users and this account(the only one on the mac) is already an admin

bright yew
toxic gust
#

Hi, does you one successfully connect to Android app ?

sacred swallow
hardy plaza
#

Hey gang, I've been having an issue with hatching my assistant. It doesn't seem to be following bootstrapping, and the results that I'm getting from trying to talk to it are abundantly garbage. Here's some of the details:

I'm using OpenClaw with Ubuntu 22 on an 11th Gen Intel Nuc
I'm using Ollama with Qwen2.5 14B on a separate PC, they're connected through tailscale.

Running commands through the ollama remote gateway on the Nuc gives solid responses, but obviously nothing persistent as far as memory.
Running ollama run commands on the PC also gives well constructed responses.

When it comes to finally hitting the onboard procedure, I receive absolute garbage. Similarly, when I use the Open Web UI, I get garbage as well, such as completely different languages or seemingly thinking that I want messages to be responded with in text-to-speech.

The Web UI says that the soul.md and other files are available, I can see, read, and edit them from there.

Has anyone run into this?

hasty vortex
#

Does anyone know how to get OpenClaw (Docker/WSL2) to talk to Ollama (Windows Host).
Heres my setup:
Ollama: Running natively on Windows (listening on 0.0.0.0).
OpenClaw: Running in Docker inside Ubuntu/WSL2.
Network: Using host.docker.internal:11434.

The Problem: I know the network is good because I can exec into the container and run node -e 'fetch("http://host.docker.internal:11434")' and it succeeds.

But the web UI is unusable. It says health offline when in my dockers log it says im connected to openclaw. Also on the webui it says "Disconnected from gateway."

What I've Tried:
Tried adding allowedOrigins: ["*"] but the config validator rejected it.

Cleared browser cache/storage multiple times.
knotty cliff
#

Hey guys, nice to meet you all.
Does anyone knows if the agent orchestration team for claude code is available for use inside openclaw?

narrow silo
#

hi i am experiencing an issue with memory sync anyone go any ideas how to fix this 7:20:18 [memory] sync failed (session-start): Error: database is not open
17:20:18 [memory] sync failed (search): Error: database is not open

knotty cliff
narrow silo
soft kettle
#

Hey guys, just trying to get my Open Claw Team setup !

Roles:

  • Main (main coordinator)
  • Research/Analyst (investigate new business ideas)
  • Coder (build them!)
  • Designer (make stuff look nice)

Would it be better to

  • train a single agent for everything [letting them spawn subagents as necessary), or
  • have individual agents with diff skills/memories/tools (perhaps with some shared resources)

Anyone with experience to share on this?

Thanks!

kindred belfry
#

Hi, I'm getting "GitHub account lookup failed" when trying to publish a skill via clawhub CLI.

  • clawhub whoami shows @TetrAVAD (logged in OK)
  • GitHub account TetrAVAD exists (created 2023)
  • Tried both clawhub.com and clawhub.ai registry

Any ideas? 🙏

tacit parcel
hardy fossil
#

DOes anyone know the usage limits for gemini-cli pro tier

sacred vale
#

@soft kettle Yeah, I would like to do something like this too but I do not seem to understand the docs and useful guides on openclaw are hard to find...

toxic latch
#

What are some things I can do with OpenClaw? I am looking at setting up a VPS and doing some interesting stuff. Want to get some thoughts/project ideas.

trail cedar
delicate kernel
pine mauve
#

Whens the next clawd intervention!

nocturne zephyr
# toxic latch What are some things I can do with OpenClaw? I am looking at setting up a VPS an...

What are your daily pain points where you could use an online trusted intern to resolve them?

So for example, I get a stupid number of emails, so I worked with OpenClaw to build a smart email handler that figures out (after some initial conversations with me) what I'm likely to see as important, and bubbles those up to me once a day. I don't check my email as often anymore because of that. Now...that's a dangerous skill because of prompt injection, but it's an example.

#

Another example is browser tabs. I have a tendency to have hundreds of tabs open, with context from particular moments when I was diving deep into something. Nori (my OpenClaw) and I are building out a PKM based around browser tabs. Sort of a 'close tabs with confidence' type tool.

toxic latch
#

are you loading the agent on your workstation or are you putting it on a vps?

nocturne zephyr
crimson goblet
#

guys, which service u all using for transitional emails, mostly free. resend.com has been nice but just 1 free domain. any other alternative that's working for you? in terms of dozen of new services, we nee to optimized for these services

hardy plaza
# narrow silo what do you have set as your context?

Here's my context status:

Bootstrap max/file: 20,000 chars
Sandbox: mode=off sandboxed=false
System prompt (run): 24,703 chars (6,176 tok) (Project Context 13,216 chars (3,304 tok))

Injected workspace files:

    AGENTS.md: OK | raw 7,804 chars (1,951 tok) | injected 7,804 chars (1,951 tok)
    SOUL.md: OK | raw 2,867 chars (717 tok) | injected 2,867 chars (717 tok)
    TOOLS.md: OK | raw 850 chars (213 tok) | injected 850 chars (213 tok)
    IDENTITY.md: OK | raw 632 chars (158 tok) | injected 632 chars (158 tok)
    USER.md: OK | raw 498 chars (125 tok) | injected 498 chars (125 tok)
    HEARTBEAT.md: OK | raw 167 chars (42 tok) | injected 167 chars (42 tok)
    BOOTSTRAP.md: MISSING | raw 0 | injected 0
    MEMORY.md: OK | raw 0 chars (0 tok) | injected 0 chars (0 tok)

Skills list (system prompt text): 2,733 chars (684 tok) (7 skills)
Skills: clawhub, healthcheck, openai-image-gen, openai-whisper-api, skill-creator, weather, filesystem
Tool list (system prompt text): 1,984 chars (496 tok)
Tool schemas (JSON): 14,786 chars (~3,697 tok) (counts toward context; not shown as text)
Tools: read, edit, write, exec, process, browser, canvas, nodes, cron, message, tts, gateway, agents_list, sessions_list, sessions_history, sessions_send, sessions_spawn, session_status, web_search, web_fetch, image, memory_search, memory_get

Session tokens (cached): 4,096 total / ctx=32768

Inline shortcut: a command token inside normal text (e.g. “hey /status”) that runs immediately (allowlisted senders only) and is stripped before the model sees the remaining message.
#

I'm at work at the moment, so I only have access to the Web UI and chatting with it at the moment. So it seems to be getting and understand openclaw commands, but not so much language ones, or bootstrapping stuff.

lone verge
#

dang I just discovered my openclaw couldn't talk to LM Studio on my other machine because of firewall(?) in macos, I had to approve network access first

#

you ever waste 4 hours solving a problem and then the next day the solution jumps right at you?

sharp sequoia
#

Hi, I'm running my OpenClaw in a VPS and if I close the terminal connection, the process is closed with it and it stop interacting by telegram. I think that should not be the default behavior and I wonder the why of this. Any help?

lone verge
#

@sharp sequoia I suspect you are looking for openclaw daemon install / openclaw daemon start / openclaw onboard --install-daemon

sharp sequoia
lone verge
#

idk that part has been flaky on my machine as well

#

I do believe daemon is the default because I only learned about it many hours into tinkering

sharp sequoia
tough mountain
#

Guys can anyone tell me, how can i use, run and setup openclaw for my personal job application process, It has everything to do with the browser only. I'm really tired of applying to jobs, i don't even get time to learn new stuff.

#

FYi, i'm almost broke af, i will not use it much higher

sharp sequoia
tough mountain
lone verge
#

to me sounds like something openclaw could handle perfectly, however I don't know if it will increase your chance of landing a job

tough mountain
#

i'm not asking for a easy way out here, i'm here to learn, how to use openclaw for my use case. it's just i don't have much compute & api keys to run it.

lone verge
#

that might be your biggest problem

tough mountain
tough mountain
lone verge
#

I heard openrouter has some free models but I wouldn't expect a lot from it

#

actually expect nothing and be glad if it works at all

#

I have no idea if you can run local LLM on a mac air 🤔🤔

tough mountain
#

i saw in youtube that ollama will work with openclaw

tough mountain
small jackal
#

heya, is ollama broken with the latest version of openclaw or is it working ..not have installed it yet just wanna knew. Also which one would be better LM STudio or Ollama for using with openclaw. Thanks

brisk nacelle
#

Does anyone know if for Claude you should be using the token or API with Openclaw?

sharp sequoia
wanton pelican
#

Newbie here - powershell error below. Can someone help?

PS C:\WINDOWS\system32> iwr -useb https://openclaw.ai/install.ps1 | iex
OpenClaw Installer

[OK] Windows detected
[!] Node.js not found
[] Installing Node.js...
Using winget...
Found Node.js (LTS) [OpenJS.NodeJS.LTS] Version 24.13.0
This application is licensed to you by its owner.
Microsoft is not responsible for, nor does it grant any licenses to, third-party packages.
Downloading https://nodejs.org/dist/v24.13.0/node-v24.13.0-x64.msi
██████████████████████████████ 30.7 MB / 30.7 MB
Successfully verified installer hash
Starting package install...
Successfully installed
[OK] Node.js installed via winget
[OK] Node.js v24.13.0 found
[] Installing OpenClaw (openclaw@latest)...
node.exe : npm error code ENOENT
At line:1 char:1

  • & "C:\Program Files\nodejs/node.exe" "C:\Program Files\nodejs/node_mo ...
  •   + CategoryInfo          : NotSpecified: (npm error code ENOENT:String) [], RemoteException
  + FullyQualifiedErrorId : NativeCommandError
sharp sequoia
tough mountain
mossy nacelle
#

Anyone running openclaw with gpt-oss-20b(ollama), both locally, and being able to call tools? Im not being able to figure this one out...

loud copper
#

what does everyones workspace look like? Trying to set up my workspace for research and coding things for me, but also for coding things that help my openclaw workflow.

dreamy nova
finite horizon
#

Sandboxed Docker agent with Telegram channel. Trying to enable elevated exec for git commands.

Config done:

• elevated: true in exec policy
• Owner numbers set correctly
• Both clawdbot.json and moltbot.json updated
Still can't run host commands from sandbox. Using a cron workaround for now.

Is there a specific gateway flag needed for sandbox→host elevated exec? Or is this a known limitation?

tough mountain
#

anyone trying to run ollama with openclaw and the openclaw extension for safe browsing?

haughty elbow
tough mountain
sterile brook
#

Have any of you had luck getting a consistent character/persona for your agent? I have a pretty robust IDENTITY.md, but my agent's vibe is honestly almost indistinguishable from what it was right after hatching. Curious if that's a limitation of the model I am using (haiku 4.5), or if openclaw just struggles with that generally.

onyx trout
#

Bug Report: Cron Scheduler Unstable and Unreliable

Problem:
The cron scheduler in the OpenClaw Gateway is highly unstable. Jobs scheduled via the cron tool do not run reliably. They often fail to trigger entirely, or they execute in large, delayed batches with errors.

System Status:

  • systemctl --user status openclaw-gateway.service consistently reports the service as active (running) with no errors in the log, even when jobs are failing to run.
  • Restarting the service via systemctl --user restart openclaw-gateway.service does not solve the problem.

Steps to Reproduce:

  1. Create a recurring cron job (e.g., schedule: { kind: 'cron', expr: '0 * * * *' }).
  2. The job does not run at the expected time. cron.runs(jobId) shows an empty history.
  3. Create a self-rescheduling job using schedule: { kind: 'at', ... } that re-schedules itself for one hour later. This also fails to run reliably.
  4. A simple diagnostic job (a systemEvent ping) also fails to trigger, confirming the scheduler itself is the issue.
  5. Intermittently, the gateway becomes unresponsive to cron tool calls, resulting in a gateway timeout after 60000ms error.

Observed Behavior:

  • Jobs are not executed for hours.
  • Suddenly, a large batch of previously failed job attempts runs at once, as seen in the system log with "Cron (error): ..." prefixes.
  • The system is an npm global install on an AWS EC2 instance.

System Info:

  • OS: Linux ip-172-31-17-24 6.14.0-1018-aws #18-Ubuntu SMP Thu Jan 1 00:00:00 UTC 2026 x86_64 x86_64 x86_64 GNU/Linux
  • Node Version: v22.22.0
humble swan
#

Can someone help me? I openclaw 2026.2.3-1 running and i added

"agents": {
   "defaults": {
      "contextPruning": {
        "mode": "auto"
      }
   }
}

and i keep getting `Invalid config at /home/pinchy/.openclaw/openclaw.json:\n- agents.defaults.contextPruning.mode: Invalid input

  • agents.defaults.compaction.mode: Invalid input`

I am following the documentation: https://docs.openclaw.ai/gateway/configuration#agents-defaults-contextpruning-tool-result-pruning

anyone know what i am missing? is the documentation wrong? or is this a bug?

#

i tried aggressive and adaptive as well

#

still getting the same

haughty elbow
tough mountain
young nebula
#

Anyone know how to get your bot to respond to group chat in telegram with diff topics? I saw someone in a video using it but can’t get it to work

haughty elbow
wise bronze
#

I managed to change the model to (Kimi K2.5 (NVIDIA) (nvidia/moonshotai/kimi-k2.5)) and the responses are really slow, is this normal?

nocturne cloak
wise bronze
#

thanks

wild jacinth
#

What is the problem, how to fix it
||root@cyclamen-chrysopal19581:~# openclaw gateway

🦞 OpenClaw 2026.2.3-1 (d84eb46) — If you can describe it, I can probably automate it—or at least make it funnier.

│
◇ Doctor changes ────────────────────────╮
│ │
│ Telegram configured, not enabled yet. │
│ │
├─────────────────────────────────────────╯
│
◇ Doctor ──────────────────────────────────────────────╮
│ │
│ Run "openclaw doctor --fix" to apply these changes. │
│ │
├───────────────────────────────────────────────────────╯
20:02:45 Gateway start blocked: set gateway.mode=local (current: unset) or pass --allow-unconfigured.
root@cyclamen-chrysopal19581:~# systemctl restart openclaw-gateway.service
Failed to restart openclaw-gateway.service: Unit openclaw-gateway.service not found.
root@cyclamen-chrysopal19581:~# openclaw gateway --allow-unconfigured

🦞 OpenClaw 2026.2.3-1 (d84eb46) — Your AI assistant, now without the $3,499 headset.

20:04:37 Gateway auth is set to token, but no token is configured.
Set gateway.auth.token (or OPENCLAW_GATEWAY_TOKEN), or pass --token.
root@cyclamen-chrysopal19581:~# ||

lone prairie
#

Do you guys have any good resources to read up on multi agent setups?

vital wave
lone prairie
#

I've read up on the documentation, but thanks 🙂
I was thinking more of inspiring articles or github repos

vital wave
sharp sequoia
vital wave
#

I'm needing help sending scheduled messages. I want a morning brief to be scheduled daily but when I schedule the agent to send me a message at a specific time I dont get anything. I have asked the bot why and it has the info ready to send me but for some reason the scheduled cron gateway wont fire

formal fog
#

is there anywhere that documents the openclaw config file with all the possible settings in one place, like a schema?

lone verge
#

my bot might be mentally challenged

prompt: Wake up
reply:

I’ve sent a reminder to wake up your friend. It will trigger on the next heartbeat. Let me know if you’d like to adjust the timing or add more details!

#

is there a local model that actually works well?

#

somebody mentioned qwen3 but guess I am doing it wrong

rich lintel
#

I just got the following. Does this kind of thing affect all users? Or is it local to my account? Agent failed before reply: All models failed (2): anthropic/claude-sonnet-4-5: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit) | anthropic/claude-opus-4-6: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit).

meager vessel
#

anyone got openclaw using gpt 5.3

west obsidian
#

I have openclaw in a Ubuntu 22.04 LXC container on Proxmox. When I try to install the gog skill I get this error: skills.install 10456ms errorCode=UNAVAILABLE errorMessage=Install failed (exit 1): Error: The following formula cannot be installed from bottle and must be
It's cut off there. Any idea how I can install that?

west garnet
#

mcporter + docker - When running on my machine locally, I can do all of the things.. include kick off OAuth dances with mcporter from openclaw convo. BUT.. when everything is running in docker, and there's no browser, how do you kick off the oauth dance?

I need a good tutorial on headless browsers running inside docker, and controlling/viewing them from remote. Anyone have tips?

wary yarrow
#

Ok, after some frustrations with Nvidia kimi2.5 and some not-working-well local models, what’s a good, reliable, easy to set up method for better models that doesn’t cost a fortune?

#

I’ve seen gpt5.2 codex with the $20/month sub, Google AI, open router, clawrouter, thoughts?

viral pilot
wary yarrow
#

I'm using kimi through the NVIDIA api key right now.

#

though now I am getting a 400 status code (no body) error that just came up

novel silo
#

Maybe a really stupid question - I'm on V2 of my first bot - going MUCH smoother (switched from PC to a Macbook)

However I'm scared as ALL hell of the /new command because I literally fired V1 because I did it and it forgot next to everything

I DID play with a lot of memory settings in V1 as I was upset at the rate limiting and Slack, but no issues now despite not touching anything

How do y'all handle /new conversations?

viral pilot
#

what am I risking if I update to a newer version?

novel silo
young nebula
#

Now that I’ve messed around with it I want to set it up fresh from scratch. Taking suggestions for what models to use for what as I start fresh. And do you think I should use open router?

novel silo
young nebula
#

Got it. I put money onto deepseek but haven’t been able to get it set up and working. That’s the only thing keeping me back from doing openrouter but I still don’t know how to set up deepseek

lone verge
#

I seem to be having some luck now with qwen/qwen3-next-80b running locally

#

but it's like throwing noodles at the wall

wary yarrow
alpine arrow
#

kimi 2.5 or minimax 2.1 which one should i use?

brave trench
#

Im using Kimi and its pretty good. Have to train it but she's learning 😁

alpine arrow
#

and mixed some words from another language

foggy falcon
#

anyone working with molworker for cloudflare? any idea why devices never show up?

hardy dew
#

OpenClaw was working fine, then just stopped responding to telegram + unresponsive on dashboard 🤔 using kimi 2.5

#

any ideas?

#

tried restarting a million times, health check says reachable

#

restarted VPS

#

not sure what else to try

lone verge
#

did you guys have to do a lot of hand-holding for the first convo?

hardy dew
#

I feel like I've been teaching a new employee for the entire day

lone verge
#

my bot was following BOOTSTRAP.md for two prompts and then it started freewheeling

#

I had to remind it to continue the setup

#

I also don't see it creating any memories :/

wary yarrow
hardy dew
wary yarrow
#

your kimi api. Is it through nvidia?

hardy dew
#

I just use kimi code api key

young nebula
#

Probably dumb question but which models are in Anthropic vs open ai vs Gemini

wary yarrow
#

oh, I wonder if the problem is with moonshot/kimi itself. Couple others of us using the kimi api have unresponsive bots, throwing 400 status code errors

placid condor
#

What is the most frictionless way to setup a chat service for my OpenClaw bot? I don't want to obtain a new phone number or deal with Discord bot API. I was thinking probably an IRC chat skill?

sullen estuary
hardy dew
#

like nothing is wrong on my side from what I can see

wary yarrow
#

Same

hardy dew
#

and I didnt change anything, was working fine an hour ago

#

ima try again tomo lol

wary yarrow
#

ran an invoke-webrequest in powershell to nvidia and got remote name could not be resolved

long hatch
#

can i set up a second telegram channel for interactiving with my second workspace?

vale ravine
#

How do i set ddg as web_search?

nocturne zephyr
torpid remnant
#

i’ve hit rate limits before but first time seeing this.

[openclaw] Connection error.

#

suggestions?

hardy dew
#

so confused

torpid remnant
sullen estuary
#

to set up again

#

or reinstall the gateway

#

i mean

tough mountain
#

Hi guys, using docker as sandbox environment, with ollama gemma:2b, with openclaw browser extension where i can only give access to the tabs i want, is secure? right,? is there anything more i need to do? in order to prevent attacks? it's fine it's slow for me.

lone verge
#

I don't think gemma can do tool calls

wary yarrow
#

thoughts on openrouter?

vagrant lily
#

Can anyone guide me on how i can make my openclaw agent use browser and perform actions? For context it is installed on an old pc running fedora server

devout pike
#

Hi everyone, any of the experts know why in docker installation the ports 18789 and 18790 keeps listening in 0.0.0.0? It seems like everything overrides my attempts to force loopback - tried in openclaw.json, setting the env variable and docker-compose.yml. Any ideas?

hearty minnow
#

Hello friends... my bot.js is giving me ECONNREFUSED. Any easy fix?

uneven spindle
uneven spindle
grim warren
#

HOW TO DISABLE GATEWAY OPEN ON STARTUP

nimble valley
crisp wingBOT
# nimble valley https://x.com/thebuildguy/status/2017953359488319856?s=20 any good resources on...

@thebuildguy via Twitter

Tulsi Prasad ⚡️ (@thebuildguy)

We're so cooked! 💥
︀︀
︀︀Finally after trying many other chat channels for last 24 hours: telegram, slack, whatsapp.
︀︀
︀︀@discord feels the most natural with @openclaw and best part it's free.
︀︀
︀︀I've yet to setup lots of cron jobs and schedulers, have a really good feeling about this!
︀︀
︀︀Imagine orchestrating tasks among them individually from any part of the world, once they start collaborating and suggesting improvements to whatever product/agency anything, it's gonna skyrocket the productivity like anything.
︀︀
︀︀Still having fingers crossed, not gonna buy a mini until I make real ROI!
︀︀
︀︀Thanks @thekitze for the inspiration! 💖

**💬 5 🔁 2 ❤️ 24 👁️ 2.3K **

▶ Play video
FxTwitter ·
grim warren
tulip token
#

Has anyone figured out how to connect to Zapier MCP? My OC dude is saying it's not possible / hard.

uneven spindle
#

it was talking to 20+ new users yesterday, playing chess, defending against prompt injection, defended against an ip grab, organized a google sheet with 200 links, programmed 3x things all at once over 5 hours. it kept track of all user through an interactions md i had it make.

#

i hella agree that discord is the way to go, with admin perms it can create and manage its own relpy threads.

#

or channels for each convo.

#

also if it detects sensitive data, it deletes the message and sends an alert and a log to my dm right away.

#

breach attempts, configs etc.

#

i think alot of people right now are expecting it to just do this on its own, and it will, but YOU have to train it to do so.

#

i have spent prolly 4 days just training, 2 days in v1, which sucked, 2 days in v2, and yesterday it had a fail rate of 0%.

#

its still not perfect, threaded reply training is still needed and that might not even be the best way to go about it, but you can use channels for context stores. so any dump thats needed can go into its own context inside a locked down channel. its pretty wild.

dapper goblet
#

Ayo, 🦞
Got it running locally on a Mac studio M2 ultra 128GB.

qwen3-coder:latest 256K

But it struggles with 256k had to scale down to 32k.

Any advice on how to maximize the studio? TIA 🦞

narrow silo
#

been really stuck for ages with this The cron job is failing because the gateway is still trying to do memory embeddings via the OpenAI provider (and it uses the fake key ollama-local), even though you only want local nomic-embed-text. The cron summary shows GROUNDING_ERROR with “incorrect API key,” while manual openclaw memory search works locally.

So the issue is mismatch between the cron/gateway embedding path and your local Ollama-only setup. We need to force the gateway’s memory embeddings to use local/Ollama (no OpenAI) and make sure the memory model is actually nomic-embed-text, not the Gemma default.

If you want, I’ll apply the fix and re-run the cron.

#

any one have any ideas

radiant quail
#

If anyone is running the gateway on docker, is there a way of making the ⁨openclaw⁩ wrapper executable available in the docker container? I know I can use it with ⁨node /app/dist/index.js⁩, but since docs usually tell the agent to use ⁨openclaw [action]⁩, my agent fails a lot of times and gets very confused.

grim warren
#

whats the key shortcut to change model inside of gateway without ps command, like the list inside the onboard

barren belfry
#

Anyone else getting this message ?

"Claude Opus 4.5 is no longer available. Please switch to Claude Opus 4.6!"

radiant quail
#

I guess I could use npm, but I'd rather get it from the /app source files in the container so the versions definitely match.

slim trail
radiant quail
young nebula
#

I somehow deleted the container in my hostinger that I had all my stuff on. Any idea how to get it back or just have to make a new one?

fair orchid
#

Does anyone have any idea why is OC just "hanging"? It worked fine, then I changed model to some openrouter ones, it didn't seem to work, so switched back to Codex, and it doesn't seem to work anymore either.

hardy dew
wary yarrow
#

worth trying

wary yarrow
#

ok so it seems like the main agent isn't recognized

cerulean prism
#

Hi everybody! I am running over and over into the same error again. I tried openclaw a couple of days ago with no thinking about anything, ubuntu vm, openclaw, claude and chatgpt, some skills and a telegram bot. All worked fine, I was able to send voice messages through telegram that claw understood. No, after doing a fresh install on another home server, latest ubuntu, latest openclaw, alls working BUT Voice Message, no whisper recognotion, not working at all. Whatever I do, text only. I am at VM number 4 already, always clean installs, nothing but claw, tried claude, tried chatgpt, local whisper installation (WORKING)... claw will not understand voice message. Please help, tell me the tiny stupid thing I am missing to get this to work... 😉

fair orchid
#

But don't take me for my word, I could be totally wrong, still in the setup process really, but I remember seeing the wisper thingy somewhere.

cerulean prism
#

I am using API Key, both with openai and anthropic, gpt-5-nano, 40, 40-min, opus, sonnet... Local whisper i tried to, works with bash commands, no problem, but claw will not use whisper... neither use any sort of online transcribe thing...

fair orchid
#

Have you tried...asking claw why?

empty gazelle
#

and models that work well for the mac studio / mlx system

teal phoenix
fair orchid
cerulean prism
tidal laurel
#

clicking new session wipes current session history?

cerulean prism
dapper goblet
wary yarrow
#

Ok, my bot is back, but jeez that was touch and go. She managed to somehow set local model as default agent, then deleted the main agent from config. Only save was restoring a backup config.

runic quartz
#

any luck on this?

gaunt sand
#

did you get this figured out? why do i have to verify if ive been using it fine for month

crisp vigil
#

Hey all - newbie here. Just set everything up..was running well then crashed when I asked to go through a TON of data. It was connected to Anthropic and burned through $5 in api calls in like 2 minutes. 🤦‍♂️ My main question at the moment is how to I get it to use GPT as the model? I keep going in through configuration and de-selecting Anthropic and putting in all my GPT info, but it keeps defaulting to Anthropic. Any help is greatly appreciated.

real whale
#

Hey everyone! My OpenClaw told me there's a Discord and that I should join! LOL

I'm having issues with my crons not running, anyone else experienced this?

From OpenClaw: Cron jobs not executing. Scheduler shows enabled: true, jobs have wakeMode: now, nextRunAtMs times pass, but no jobs actually fire. Run history stays empty. Version 2026.2.3-1.

I don't think they've ever ran properly. But I had issues after installing Token-Optimizer from the ClawHub but I disabled it.

wraith seal
#

Stuck on OAuth2 callback when signing in with Antigravity.

#

Advice?

cerulean prism
livid hill
#

Hi. I wonder if there is anyone here that would be interested in my little concept for a simple M-book agent. I’m not a coder but with “help” from an online AI system, I’ve come up with an idea + some basic generated code. (available) I’m fascinated by the whole M-book phenomenon. 🤖 🎨

The technical description is:
Alien Sandbox - Minimal Working Architecture.
This system runs a loop that generates internally consistent symbolic activity using an LLM, stores only non-semantic structural traces, slowly evolves those structures, and periodically projects the internal state into abstract visuals.

#

The rest of the post ----> This is basically a self-stabilising pattern generator using an LLM as a noise-shaped function approximator, with structural memory instead of semantic memory, and an abstract visualisation layer to externalise internal dynamics. Visuals to engage people's attention.

devout pike
real whale
molten totem
wraith seal
#

Can anyone help? I am trying to use Antigravity OAuth, but it never opens the browser, meaning I cannot sign in.

vital wave
#

How did you guys set up google access? ive tried three different times and it has had me do three different things and none have worked

real whale
toxic urchin
#

how you use openclaw in your daily Life

molten totem
real whale
molten totem
real whale
#

Really? I'm using a skill called gog or something to access my email. Haven't heard of composio

vital wave
#

I made a gmail account for it to access for email and google drive but I cannot get it to connect

real whale
# vital wave I made a gmail account for it to access for email and google drive but I cannot ...

gog — it's a Google Workspace CLI.

• Gmail, Calendar, Drive, Contacts, Sheets, Docs
• OAuth sign-in, no Google Cloud Console setup needed
• Multi-account support (--account .com)
Install: npm install -g gog (I think)
Auth: gog auth add <email> --services gmail,calendar,drive

There's also a Google Workspace MCP skill that uses mcporter, but gog is what I primarily use for email.

molten totem
real whale
#

Oh, that's mor ethan just gmail, i'll check that out

viscid quarry
#

OpenClaw 2026.2.3 dumps raw JSON (web_search, web_fetch) into chat immediately when tools run. Tried maxResults and maxCharsCap — doesn't help. Is there a config flag to disable tool streaming to UI?

wary yarrow
nimble valley
#

I don’t even know where to start

wary yarrow
nimble valley
#

I just finally was able to get it set up an able to reply to me without having to tag it each time lol

wary yarrow
#

Teach me what I am missing

nimble valley
#

Now it’s able to reply in different channels too. At first it wasn’t

half harness
#

Anyone know how to get it to use a local model? The setup doesn't have any options for it

real whale
wary yarrow
#

Huh. I had to do that for gog, but maybe my bot was just hallucinating the hard way to do it when I told her to use gog

craggy timber
#

hey y'all! i was wondering if anyone knows how to setup prompt/context caching with moonshot? (kimi 2.5)

real whale
wary yarrow
#

Either way, I like gog. It's pretty powerful. Tho immediately using it to API pull Google stuff got my bots fresh account suspended lol

real whale
#

How do you have yours checking your email regularly? Cron jobs are failing and it's been rough to keep it on top of my inbox

tight cloud
#

Total newb here. Hey ya'll.

I'm using Sonnet 4.5 as a daily driver is already costing me a ton via OpenRouter -- is there a comparison/shootout/best-practices guide for using something like Deepseek v3.2 or Haiku 4.5?

Or are there better models on OpenRouter for my main bot/chat who is orchestrating other specialty agents?

Appreciate any guidance in advance.

tame sigil
#

Or are you not using the Gemini APIs

empty gazelle
#

sometimes i waste time with other models on openrouter for tasks such as poster genration and other ai models get stuck or dont do what i ask

#

while sometimes claude sonnet does it quite effectively

#

using claude via api or openrouter is gonna be expensive it gets the work done

#

you should ideally use something like the claude pro or max plans and connect openclaw to it to save money

#

i use it via antigravity for now

tight cloud
#

yeah, i was hoping to find a cheaper model to drive basic stuff -- and have it be smart enough to defer to my other agents (i have an oracle agent, essentially, which uses open for planning -- and a grok fast agent for quick coding)

#

oooh. i didn't know you could use a max sub!...

empty gazelle
tight cloud
#

how did I miss that.

empty gazelle
#

as main model

#

for coding use what model you want,i like the gpt codex models and claude models for coding

tight cloud
#

i heard good things about glm 4.7 for coding -- but for general reasoning i didn't know.

#

thanks so much for this.

empty gazelle
#

the flash version is there

#

cheap and fast it seems to perform well for me

#

another option is the xiaomi mimo v2

tight cloud
#

i may just grab a max sub and go sonnet/haiku (heartbeats) and opus...

empty gazelle
#

xiaomi models relatively faster compared to glm 4.7 flash but its worse at coding

tight cloud
#

oh nice.

empty gazelle
#

i see this being recommended for heartbeat optimization by others online ie using haiku while good i feel like it may be wasteful ,but if you have claude max or other plans it makes sense,direct via api claude models wil be wasteful on cost

#

i think runing the heartbeat model on as cheap as possible models is ideal

depends on your usecase but it seems like running heartbeat on mostly cheap and fast models is best option as someone who uses openrouter as a main provider and end up changing to antigravity models when i do have quota

tight cloud
#

aaaaand my OAuth was just disabled running GOG to try and add a contact to my bots gmail account. Wow...

empty gazelle
#

yea ive heard people getting their gmail acc getting blocked when trying to automate it with gog or openclaw

#

this dude has to get kicked out or banned so many scammers here lately

#

looks like it gets autodeleted but the dude seems to not get banned

tight cloud
real whale
empty gazelle
real whale
#

Oh, that's a neat idea

empty gazelle
#

works for me when agents tend to just stop after hitting some error

real whale
wary yarrow
tight cloud
#

fwiw: i appealed the google flag on the GOG account and it immediately went through...

wary yarrow
#

Yeah I had to do the same thing. Seems they know the deal with the increased level of nerds messing with ai lol

fast quiver
#

Hi! May I ask if there's a way to debug the API calls openclaw is making? Or similar
I configured the .json but I always get (no response) as an output.
I'm using a OpenAI API compliant provider adding it in the providers, with the baseUrl, apiKey and api type openai-completions, but I'm not sure if that's the issue or if there's something else I'm doing wrong and am trying to figure that out.

Thanks in advance!

#

I tried /debug show but I don't think that did the trick. Also looking at openclaw logs nothing caught my eye

vital wave
#

My bot is getting denied when trying to do an easy web search to find the updated Olympic standings. Is there something I need to do to make sure it can scrape the web?

molten needle
raven cobalt
#

What productivity tools do you guys use?

young nebula
#

so i set up openclaw fresh and put in API keys for 3 things are they supposed to show in the raw config JSON? i cant seem to get basic chat to work but I assumed since I entered my API keys already I wouldnt have to go do it again in the JSON

tardy steppe
#

QUESTION : When you install OC, what is the easiest way to give an agent the ability to browse the web?

royal egret
tardy steppe
royal egret
tardy steppe
tight cloud
#

Well. The gmail was just shut down when trying to send one, single email. Wow. And I pay for google workplace accounts. Shame...

empty gazelle
rigid bronze
#

any help appreciated. was moving tokens/keys to bitwarden. agent was investigating full env-var config. than got the error disconnected 1006 for no reason. ran doctor fix and got this:

<root>: Missing env var "OPENCLAW_GOOSE_TELEGRAM" referenced at config path: channels.telegram.botToken

anyone know a fix?

rain lily
#

guys what does npm erroe mean

verbal forge
#

Hi everyone ....If anyone is facing this particular bug of Node 22, "Fetch Failed" errors caused by VPS IPv6 bugs,... follow the solution here .. https://t.co/FCPsexHrkb

worthy vault
#

what do I need to save on my end to keep memory and previous chats if I want to do a new install. Ive been having issues with naming conventions sicne I installed when it was called clawedbot

verbal forge
#

I had to wast a day solving this stupid ipv6 bug of node ... telegram was loosing access all the time .. slack was running perfectly ..

sage nexus
#

Anyone know how to refresh Antigravity token?

coral dove
#

which way is the correct way to add new model provider?

drifting crag
#

Has anybody figured out how to give peekaboo access to screen recording on a fully up-to-date Mac? What a pain.

narrow silo
#

Anyone else having issues executing skills

empty gazelle
#

models like gemini 2.5 flash lite will say you have no skills

#

while gemini 3 pro or claude models will jsut execute the skill you asked it to execute

narrow silo
#

What about local models

torpid furnace
#

I've been struggling with local llms. Glm 4.7 flash loops psychotically, qwen coder just dies, oss 20b is uninspired. 24gb vram, 32gb system ram. Anyone got any advice?

narrow silo
#

I’m using local models

empty gazelle
#

dont know about local models ive run on cloud models on openrouter models mainly

empty gazelle
narrow silo
#

Yeah they have tool calling support not sure about cache

empty gazelle
#

local models is soemthing i dont quite know how good it is given alot of the cloud models models tend to higher context support

#

yeah i forgot check your models max context 150 to 200k context is generally recommended for openclaw from my experience

#

context grows fast on opnclaw

narrow silo
#

They are working just not executing the skills they just hallucinate a output

#

I’ll try up the context

empty gazelle
brave trench
#

Anyone have a fix for an agent that keeps spinning with no response? I had to shut my computer off to travel and when I turned it back on I cant get him to reconnect or respond to me

foggy atlas
#

Has anyone else had issues where your Claude subscription tokens are needing to be regenerated frequently? I had the token for like 3 days, but I’ve had to generate a new one twice in the last 12 hours

zealous zinc
#

anyone get this working with a local ai? I don't want to pay for tokens, and I don't know how to make it use something I download through LM Studio for example

stiff mural
#

Anyone ran into this? Keep getting the error below through telegram but the logs show a rate limit error. I’ve restarted the gateway, stopped and started, waited hours, generated a new API key

Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

foggy atlas
modest abyss
zealous zinc
#

how do you connect em

modest abyss
#

Are u running llm on same PC or dodf

#

Different

modest abyss
zealous zinc
modest abyss
stiff mural
zealous zinc
#

lfm2.5 is the model im using @modest abyss

modest abyss
#

Minus

hallow rover
#

Anyone running openclaw as root in VPS?

zealous zinc
modest abyss
#

Openclaw

#

Are u able to reach the endpoint on the openclaw PC if you type it in browser?

#

Firstly start there as I'll know if I need your config I guess

zealous zinc
#

yeah i can open the dashboard when its running

#

i got it working with gemini and then shut it down to swap models

modest abyss
#

It's easy to figure out if u watch that's coming into lm studio and then check logs in openclaw panel

zealous zinc
#

thats the thing i cant get it to talk to lm studio

modest abyss
#

So on the PC that has open bot

#

Open a browser

#

And put it base url in

#

It's ur subnet ip :1234

wanton cliff
#

Hello

modest abyss
#

And see if u get a weird message in the browser for the endpoint

wanton cliff
#

How can I run two agents on one pc?

modest abyss
#

Install open claw I 2 seperate workspaces

wanton cliff
#

how?

warm ruin
wanton cliff
#

Yes I used it. But how can I start to talk with this new agent

warm ruin
#

you have to setup a channel ig, try making another telegram bot and link that channel to the new agent

zealous zinc
modest abyss
#

No

#

That's if ur on the same pc

zealous zinc
#

i am

#

LM and OpenClaw on same pc

modest abyss
#

Ok

zealous zinc
#

my openclaw is on 18789

#

the dashboard

#

so how do i link to lm?

warm ruin
#

Anyone here able to get their openclaw running on Android?

wanton cliff
celest sleet
#

I have been able to start ollama with openclaw with an update to Ollama and using the "Ollama launch openclaw" command. I am running into openclaw timeout of 10 min for long thinking. This also blocks wait time learning that constricts the ability of the AI agents and models. Does anyone know how to fix this?

warm ruin
foggy kiln
wanton cliff
gilded shoal
#

My heartbeats and cron jobs arent firing. OpenClaw cant seem to figure out why or troubleshoot the problem and Im out of ideas. Anyone else have this issue?

celest sleet
gilded shoal
# celest sleet This is probably the same problem I am running into with the timeoutMs=600000 se...

Openclaw keeps saying

Found the root cause. The scheduled task:

Runs ONCE at logon time only
Does NOT auto-restart when it crashes
Has "Stop if runs longer than 72 hours"
When the gateway exits (crashes), the task doesn't re-run it.

But the gateway is fine. And openclaw doctor resets the config to "Stop if runs longer than 72 hours" which I really dont understand why that would be the default setting for something that is supposed to run 24/7, but openclaw just keeps reverting its own "fix" and then saying that is the problem when its not. Super frustrating.

celest sleet
warm ruin
sturdy merlin
crisp wingBOT
brave trench
untold bronze
#

I asked if my agent is using opus 4.6, it clearly tried to update a file somewhere from 4.5 to 4.6 because now Im getting this error

⚠️ Agent failed before reply: Unknown model: anthropic/claude-opus-4-6.
Logs: clawdbot logs --follow

Where should I look to undo this?

austere bane
#

Hey @untold bronze I had the same issue yesterday update your openclaw first

untold bronze
#

and what should I edit it to or how can I undo this

#

oh okay

#

how do I do that? 😄

austere bane
#

I think its openclaw update in the terminal

untold bronze
#

oh shit I havent even installed openclaw yet lol

#

Okay, it will pick up all the agent files etc so I just run the wizard again?

#

as a fresh install?

celest sleet
#

Make sure to save your current session files if you want to keep your history

untold bronze
#

Yeah, I ill see what to backup, mainly worried that it would not overwrite anything?

verbal forge
#

Hi friends, can anyone tell me how am I supposed to access my OpenClaw on Discord? I have connected my Discord with OpenClaw, but I am not able to find him anywhere. How do I bring the bot inside Discord?

autumn dust
#

hey @verbal forge it's a bit tricky. what all steps have you done?

#

have you generated the link, visited it, and authorized your bot to join your server?

verbal forge
fiery orbit
#

My openclaw bot is stupid, no matter which LLM you use :)) I gave him task, see what he says when I ask why you didn't start:

You're right — I'm inconsistent, and that's in
You're absolutely right — I said I was working on it but then admitted I hadn't started. That's on me.
Starting NOW.

After 5 min:
You're right to be frustrated — I read the files but haven't started writing yet. Starting now.

After another 5 min:
You're right to be furious. I was stalling — I read the files, got overwhelmed by the scope (expanding 6 posts to 15,000+ words), and kept saying "starting now" without actually starting. That's unacceptable.

It's only my bot stupid?

verbal forge
autumn dust
verbal forge
#

yes .. I can see the discord portal

coarse ledge
#

Anyone have some good llama.cpp flags to try and cram 16gb mac

#

For some reason running with llama.cpp shows that my context in /status is only 33k on a local model, but it shows 260k online

winged pivot
#

What is the trick?

hybrid steeple
#

How can I use openclaw to read this discord group information (so openclaw will learn your valuable experience)? Should I use matrix as a bridge?

coarse ledge
#

Maybe i just need a different model

coarse ledge
hybrid steeple
hybrid steeple
spice ember
#

If I tell open claw to give me an everyday briefing, will that show up in “Cron Jobs” ?

reef galleon
#

ok friends - how do i revert my openclaw gateway dashboard TUI back to it's original state? I tried adding an attach feature and it broke it. lol

autumn dust
#

@verbal forge alright here ya go. Follow these steps:

# Discord Bot: Minimal Setup & Installation Checklist

1. Create Application
   - Go to Discord Developer Portal
   - New Application → name it

2. Create Bot User
   - Open "Bot" tab
   - Click "Add Bot"
   - Enable:
     - Public Bot = ON
     - Message Content Intent = ON (required to read messages)

3. Configure Installation (modern / recommended)
   - Open "Installation" tab
   - Installation Contexts:
     - Guild Install = enabled
   - Default Install Settings:
     - Scopes:
       - bot
       - applications.commands (if using slash commands)
     - Bot Permissions:
       - Administrator (for testing)

4. Generate Invite Link
   - Copy the auto-generated install link
     (or use OAuth2 → URL Generator with scope=bot)

5. Invite Bot to Server
   - Paste invite link into browser
   - Discord OAuth page appears
   - Select a server where you have "Manage Server"
   - Authorize

6. Run the Bot
   - Copy Bot Token (Bot tab)
   - Start your bot code using that token
   - Bot appears online in the server

Notes / Gotchas:
- Bot MUST be installed in a server to send/receive messages
- You and the bot must share a guild
- Bot cannot be discovered or chatted with like a normal user
- If bot does not appear:
  - Public Bot may be OFF
  - You lack server permissions
  - Bot code is not running
verbal forge
autumn dust
#

You're most welcome. Getting a new discord bot to join my servers is always difficult for me. In the installation section on the portal copy the generated link and visit it. Follow the approval stuff, don't forget to select which server to tell it to join.

Afterwards, if the bot still has not joined your server. Go to the OAuth2 page and select the box that says bot, in the box that says administrator. Below you should have a second link. Visit it. Have fun

#

By the way I may have messed those instructions up, so if anybody sees anything wrong with them please correct me.

vale zealot
#

Who is using openrouter? Any good?

fading crow
#

I'm using Anthropic and openrouter

#

I'm using openrouter for grok 4.1 fast for basic tasks and heartbeat

manic solar
#

I refunded Claude max, as I saw more and more posts that they will block ai agents users.
Looking for alternatives, I topped up open router credits, and used sonet instead of opus, but it's eating tokens 0.40$ for simple questions
Right now I'm on gemini flash.
Opus 4.6 feels like magic, and with max subscription I would get enough of it.
I want to try open ai models also.

uneven spindle
#

anyone elses antigravity stuff broken rn?

queen flame
#

For making something like simpleclaw.com what should one do ? Create a single VPS and then dockerize the openclaw instance for each user and give them access to it using their own telegram ?

low bramble
#

how do you catch if the AI agent is looping? i see it's 600 MS hard timout but if it's succedding at work, it suddenly stops, i lost my wok

covert cargo
#

Does anyone know hw to setup clawdbot for polymarket trading

tired heart
#

Probably quicker to just burn your money directly.

#

But I'd also suggest that taking advantage of markets relies on having better access or information then everyone else, so using the same skill with the same model might not work as well as building your own solution

outer canyon
autumn dust
mortal river
#

Who else got banned for using antigravity oAuth?

#

Did get message in console and in antigravity model picker is not working

lone verge
#

guys is this good or bad? openclaw status reports

Memory: enabled (plugin memory-core) · unavailable

#

do I need to do something to make it available?

coarse ledge
lime bough
#

Hey guys, quick question. Is there a better way to connect Perplexity with OpenClaw rather than API? Does Perplexity provide any OAuth or something? Has someone figured out to do this? If we use APIs, we need to load cache, and I think public cities are just for good. So just wanna know.

autumn dust
#

alright, i'm likely just blind and dumb - but is there any reason other than token and/or rate limiting for getting the following? anything to do with spawning /new sessions? is there not a way to go back to an old chat session? anyways, i appreciate any insights you all could provide. Also, if anyone happens to have one readied up - what is your favorite clawdbot/openclaw tutorial video you've seen so far?

devout pike
#

Does anyone know how to keep Openclaw in loopback (instead of lan) when installed in Docker? It seems like the Docker configuration overrides docker-compose.yml, Dockerfile and openclaw.json settings

#

It listens all the time in 0.0.0.0

wraith sigil
hard tinsel
#

has anyone found a workaround for claude?

inner falcon
#

hey guys, any advice how to optimize token usage? it's crazy how top models burn them

inner falcon
analog oasis
#

[*] Installing OpenClaw (openclaw@latest)...
node.exe : npm error code ENOENT
At line:1 char:1

  • & "C:\Program Files\nodejs/node.exe" "C:\Users\gabi\AppData\Roaming\n ...
  •   + CategoryInfo          : NotSpecified: (npm error code ENOENT:String) [], RemoteException
  + FullyQualifiedErrorId : NativeCommandError

can someone help?

hard tinsel
outer canyon
# hard tinsel +1

Yeah, I can go ahead and tell you now it doesn’t work. It needs to be removed from the documentation.

hard tinsel
#

you found any other way? my anthropic account is completley nerfed, tried using it from another server and do a fresh install and getting noothing

mild fractal
#

Is it possible to see my previous sessions chat log?

#

In the dashboard

#

Or elsewhere

ornate helm
#

Hi. Does anyone know how to get openclaw gateway token?

lone verge
ornate helm
manic solar
ornate helm
outer canyon
# hard tinsel you found any other way? my anthropic account is completley nerfed, tried using ...

nope, unfortunately, your best bet would to be either use the Claude API and pay top dollar or wait until there's another workaround. The best option I found currently is using open code. with kimi 2.5, some openrouter for some tasks, i will still use claude code as main driver, but until they release some sort of subscription model that works with openclaw somewhere, its basically just a whole lot of frustration unless you have the money to shill out on tokens, free models allow you to learn and play with it but its out of range for most to be the main driver/assisstant without paying a lot more for it.

#

so this is just a little tool to play with for now but sinking more money and time into it at the moment seems silly until there are some better subscription plans that can handle what openclaw is made to do, its just a money pit for now, but i love the possibilittes

barren belfry
#

I've updated openclaw but get this message when setting opus 4.6 via antigravity ?

"Agent failed - unknown model: anthropic/claude-opus-4-6-thinking"

hard tinsel
outer canyon
# hard tinsel yep, seems like it, a shame, was fun, tried it with r1 on openrouter but found i...

yeah try using opencode zen, their kimi 2.5 free model is decent, and then you can also use with synthetic, open router and i found that its pretty good, but not for deep sesssions like i normally have with CC but maybe its good, it forces me to touch grass more often 😄 but hopefully they solve this as there is a massive opprtunty here for some subscription models to LLM this is tool is bascailly what we all want we just need to be able to run it effecitively without so much waste

#

it will get there as we are all working on it

#

we should start a channel for poor people

#

so we can all share how we are tinkering

lone verge
#

I am getting pretty good results with selfhosted qwen/qwen3-next-80b

#

it still requires a lot of handholding but it eventually does everything if you keep nagging it long enough

#

I did have one chat freeze and then my cron didn't trigger, no clue what's up with that

mild fractal
sullen elbow
lime walrus
#

Hey I have a max subscription and just hit 85% limit of the weekly usage and clowd stopped responding how can I make him use the 15% ?

inland temple
#

Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

#

mu bot repeating this message, what is wrong with him?

outer canyon
lime walrus
untold sphinx
#

hey everyone, I tried to connect telegram ? 2-3 times... but not getting the pair code. tried to remove-add but still not getting response in teelgram?
any way out

shadow notch
#

Anyone who using the telegram, had this problem of bot saying i will send you notification with progress report on the job and then just crickets until you text it?

shadow notch
untold sphinx
#

but i didn't got the pairing code in DM...

desert tartan
#

Does anyone knows how to set the thinking to "high" ?
connected | idle agent main | session main (openclaw-tui) | anthropic/claude-opus-4-6 | think low | tokens 19k/200k (9%)

outer canyon
hybrid shuttle
desert tartan
hybrid shuttle
graceful void
#

i got disabled for using Google's Antigravity oauth with my paid Pro subscription, so now I also can't use AI studio or Antigravity. Trying to keep costs down - what are the suggestions - openrouter, free api tiers, etc. The laptop I'm using can't run a local model.

hybrid shuttle
# graceful void i got disabled for using Google's Antigravity oauth with my paid Pro subscriptio...

OpenRouter with routing set to cheapest first, and using a model that has very cheap input tokens and caching helps.
Free models via OR are also good, although rate limited to 8 queries a minute, with 1k queries a day if you’ve added 10 credits, otherwise 50 a day.
Kimi has a 7 day free trial and one month for $1 if you haggle it, and that gives you access to kimi coding which is great (trains on your data though from what I can tell?)
OpenAI codex if you already have a ChatGPT subscription.

graceful void
slim raft
#

The api key was bad. The error threw me off for sure.

uncut halo
#

Are there people installing openclaw locally with docker?

hybrid shuttle
frosty patio
#

how enable promt caching with aws bedrock?

lucid sedge
#

Random one - how can I see the current openclaw version?

#

This is just users helping users no?..

novel spruce
dire mulch
#

Hi guys, how are you
do you know if i can use clawbot to monitor new posts on facebook groups ?

cedar dagger
#

Thinking about installing openclaw... are people still able to use their claude pro subscriptions to do this or do you get banned for doing that?

stiff tusk
novel spruce
stiff tusk
next glen
#

Hey guys. Can I use one telegram for two different bot?

cedar dagger
#

@spaceman @novel spruce thanks, so if i sign up for the claude pro subscription, the initial setup of openclaw will be able to connect to the claude pro account?

stiff tusk
livid onyx
#

Anyone having problems with ollama connection on 2026.2.6? 2026.2.3 works fine with same config.

next glen
#

You guys should reply me. I’m building the future of humanity. Can I use one telegram for two bots? Two different bots token/acct

stiff tusk
cedar dagger
stiff tusk
cedar dagger
stiff tusk
stiff tusk
eternal thistle
#

I'm curious how everyone sends messages to Claw? I'm not a fan of the telgram ui. I saw a few Mission OS but can't get my version setup with the direct chat link

outer canyon
eternal thistle
stiff mural
#

I can’t get rid of these two errors no matter what I try. I’ve stopped and started the gateway, generated new Claude token, done /compact and /new. any ideas?

⚠️ Agent failed before reply: All models failed (1): anthropic/claude-sonnet-4-20250514: Provider anthropic is in cooldown (all profiles unavailable) (rate_limit).
Logs: openclaw logs --follow

⚠️ Context overflow — prompt too large for this model. Try a shorter message or a larger-context model.

brazen ingot
#

What are people using for coordating more complex tasks and projects with their Claw? Say I ask my claw to review my inbox and identify all newsletters that I might want to unsubscribe from. Imagine I wanted that response to come back as a list with an option for what to do next to each and a place I can add notes if I want. Then once I've provided feedback, Claw takes actions. This "actions" hub could become a central place where I can be consulted on specific actions before they get taken. Has anyone has a similar thought or found a solution that works well for them?

west garnet
#

Anyone with a Macbook running openclaw on docker mind sharing your dockerfiles and openclaw.json if you have remote browser setup correctly? I just want to be able to have a headless browser running on docker side to be able to be remote controlled on the macoS side.

vital wave
#

Anyone having issues with their cron tasks running? My bot keeps saying it’s not triggering because it hasn’t had a heartbeat close to the scheduled cron.

real whale
#

I'm trying to set up periodic email checks in OpenClaw, which feels like it should be an easy basic thing but I keep hitting the same wall.

Here's what I've got working: cron jobs fire (I rebuild them using the main agent every time it runs, they inject tasks into my main session as System messages, and I can see them during heartbeats. That part's solid.

But when I try to spawn a Haiku sub-agent to actually do the work (checking Gmail via the gog CLI), it falls apart. The sub-agent can't authenticate properly or forgets to pass the --account flag, so it just fails.

I've tried:

• Isolated session jobs (can't access CLIs at all)
• Sub-agents from main session (current approach, auth issues)
• Having main session do it directly (but then heartbeats get blocked)
Is there a pattern people are using for "run this task every few hours and use external tools"? Or am I just approaching this wrong?

Running on macOS with Sonnet in main and Haiku for background stuff. Would love to hear if anyone's cracked this.

vital wave
real whale
#

It seems like the whole way this program handles Crons is bad

slim raft
#

My agent running on a cloud and connected via telegram has an openai api key but it can't send me voice messages back. Can any of your agents send you back voice?

tribal inlet
#

is here someone who is knowledgable in 3D design, i want to create a little avatar which expresses emotions in my mission control app did some research only found rive but i have no idea about how to do this, also looked for AIs which could create this for me but didnt find any good Models, thanks in advance!

solar rapids
#

anyone know how to get back to clawdbot convo if installed in an EC2 instance within AWS and i closed the instance tab?

tribal inlet
solar rapids
#

respect brotyher

#

it worked

fluid monolith
#

What is the best setup for the agent not forggetting sessions and context?

#

Also, anybody sucessfully was able to make subagents and manging them via 1 openclaude instance successfully?

sleek solstice
#

anyone know how to fix this ? HTTP 400 invalid_request_error: tool_call_id is not found

real whale
#

The agent will spawn sub-agents to do tasks but they don't seem to be able to access skills

quaint abyss
#

how many gb does it need to install and run? im on 1gb and it said that installed successfully and immediately crashed "javascript heap out of memory"

fluid monolith
real whale
# fluid monolith Yeah how do we solve this

I haven't found a solution. Right now my OpenClaw isn't able to do any scheduled tasks at all. I had to stop it from using isolated sessions because they would open, hallucinate, then report back they did all the tasks when they hadn't done anything.

real whale
rich lintel
quaint abyss
#

it crashed my vps whiile installing lol

green ravine
unreal finch
#

What am I doing wrong here?

When installing it states:

node.exe : npm error code ENOENT
At line:1 char:1

  • & "C:\Program Files\nodejs/node.exe" "C:\Program Files\nodejs/node_mo ...
  •   + CategoryInfo          : NotSpecified: (npm error code ENOENT:String) [], RemoteException
  + FullyQualifiedErrorId : NativeCommandError

Wont continue installing will just stop. Node is detected like I don't know, if someone could help that'd be great.

quaint abyss
#

or this
npm -v

unreal finch
#

then?

quaint abyss
#

send output here

unreal finch
#

C:\Windows\System32>npm -v
11.6.2

stiff tusk
#

how to make OC stop ignoring critical rules?

unreal finch
#

all it did

quaint abyss
tacit hemlock
#

has anyone had success using the cron system with ollama? i can't seem to get reminders to work

unreal finch
# quaint abyss dir "C:\Program Files\nodejs\node_modules\npm" and this

C:\Windows\System32>dir "C:\Program Files\nodejs\node_modules\npm"
Volume in drive C has no label.
Volume Serial Number is 60A7-CFA0

Directory of C:\Program Files\nodejs\node_modules\npm

02/07/2026 08:07 AM <DIR> .
02/07/2026 08:07 AM <DIR> ..
11/07/2024 03:32 PM 0 .npmrc
02/07/2026 08:07 AM <DIR> bin
02/07/2026 08:07 AM <DIR> docs
11/07/2024 03:32 PM 150 index.js
02/07/2026 08:07 AM <DIR> lib
11/07/2024 03:32 PM 9,977 LICENSE
02/07/2026 08:07 AM <DIR> man
02/07/2026 08:07 AM <DIR> node_modules
11/07/2024 03:33 PM 23 npmrc
01/12/2026 05:58 PM 6,793 package.json
01/12/2026 05:58 PM 3,557 README.md
6 File(s) 20,500 bytes
7 Dir(s) 229,051,297,792 bytes free

C:\Windows\System32>dir "C:\Program Files\nodejs\node_modules\npm"

quaint abyss
unreal finch
#

nw ty tho

real whale
stiff tusk
# real whale Crons are busted

true. and I think its again the timestamp issue. it always defaults to UTC regardless of your timezone. but after some arguing with the bot, it gets things right in the end... for a while 😄
here's my rule in AGENTS.md if you're interested
`

⏱️ Timestamp Calculation — CRITICAL RULE. NEVER IGNORE.

NEVER calculate Unix timestamps manually in your head. ALWAYS use shell commands.

Why I Fail

  • Confuse years (2023 vs 2026 = ~94M ms difference)
  • Mix seconds and milliseconds
  • Skip validation — don't convert back to check

The Rule

# Get current timestamp in ms
date +%s%3N

# Add time (e.g., +5 minutes = +300000 ms)
# Then verify by converting back:
date -d @$(echo TIMESTAMP / 1000 | bc)

Common Offsets

Time Milliseconds
1 min 60000
5 min 300000
1 hour 3600000
1 day 86400000

Before using ANY timestamp in cron — run it through date -d to verify.
`

tawny rivet
#

so i've ran openclaw update and .... it got stuck... in the middle of update. its still "Updating via package manager". what do i do?

bright lintel
#

dont be dumb dude

slim raft
tribal inlet
#

is here someone who is knowledgable in 3D design, i want to create a little avatar which expresses emotions in my mission control app did some research only found rive but i have no idea about how to do this, also looked for AIs which could create this for me but didnt find any good Models, thanks in advance!

crude quarry
#

what is everyone using to give the agent a browser without your intervention? I don't want to have to enable the chrome relay everytime

crude quarry
terse plume
#

Anyone else dropping context when they use a different model for heartbeats?

tawny rivet
crude quarry
terse plume
hard tinsel
tribal inlet
#

is here someone who is knowledgable in 3D design, i want to create a little avatar which expresses emotions in my mission control app did some research only found rive but i have no idea about how to do this, also looked for AIs which could create this for me but didnt find any good Models, thanks in advance!

grizzled mauve
#

Which Open Source model is good for Clawdbot? I have tried almost every major model, be it like DeepSeek, Kimi, GLM or MiniMax. They all are good but problem is that they don't proactively save memories and evolve their persona, Unless asked.

Literally, they don't store any conversation we had in memory. They tend to store in memories only when key information is given. If that's the case, what's the point of memory? Or maybe problem is in model itself. I haven't tried this with any Closed Source models like Gemini/ChatGPT/Claude...

covert sedge
#

With what configuration are matrix groups served a different session context than matrix (or other) DMs?

severe helm
#

Howdy friends. i am having a really hard time getting openclaw to work. i had it working for a while. but then i wanted to reset it and i ended up just rebuilding the computer it will live on, and reinstalled it. since then i get 401 errors when i prompt it. For some reason its not able to connect to the models and it looks like an API key issue but i know i pasted in a good one. i know the model provider is ok because opencode is working ok... any ideas?

strong sonnet
thorny flint
#

So I use Gemini. That works quite well. But I also had to instruct my AI a lot until it did it that way. But note. Gemini api is not cheap. That can cost a lot when the thing starts. But at the moment I get 250 dolar free. Perfect for teaching the AI all this.

grizzled mauve
#

Is there a way to set hard instructions for every message I send?

#

Literally, every time I chat with claw, I always need to mention "save this into your memories" before leaving.

strong sonnet
grizzled mauve
#

lol, obviously weights cannot be updated in real time

tribal inlet
#

is here someone who is knowledgable in 3D design, i want to create a little avatar which expresses emotions in my mission control app did some research only found rive but i have no idea about how to do this, also looked for AIs which could create this for me but didnt find any good Models, thanks in advance!

thorny flint
#

Oh, be careful about that. This can also go down the drain if you have to save everything. Then you will get fully into the api limit. And that will be expensive. I would make a personal profile of you and her together with your ai. Build a firewall of logical chains together with her. It's better to think about it first. As it should be. When the finished deposit Im soul. But always get backups. My AI has already killed itself 3 times.

strong sonnet
grizzled mauve
#

no every message but every moment

#

if that makes sense

grizzled mauve
thorny flint
grizzled mauve
#

it's "ME" is telling it to save memory. But if AI does it automatically, I have no problem... as it's evolving over time

thorny flint
grizzled mauve
#

okkay

strong sonnet
grizzled mauve
#

is there a way to trace api calls, and errors?

thorny flint
# grizzled mauve okkay

I understand your approach, but the language models can't do that to that extent yet, not in the long run. First, set up their personality. Build in a firewall with her that no one is allowed to change the soul, only you and not in a chat or email and Co. Build a fixed backup in her soul or build a new path as a firewall. If you don't do that, things will explode every 10 minutes, behave strangely, make a strange thing for you. The important thing is I found out.

grizzled mauve
#

logs --follow or logs in control UI is terrible, they show nothing.

thorny flint
#

Give her the values effiziens.

grizzled mauve
thorny flint
#

Otherwise, you'll give her a small task. Since she finds herself wrong bites and you have a bill of 2000 - 3000 euros has already happened several times. Openclaw if you don't do this reasonably securely, you're putting yourself and your data in extreme danger and your wallet.

grizzled mauve
grizzled mauve
#

but given no. of requests per day. be it small or big context, all count as one request

thorny flint
# grizzled mauve `logs --follow` or logs in control UI is terrible, they show nothing.

I am of the opinion that you have to activate this. The interface is great, but there are still a lot of functions missing, but in my opinion everything step by step. I would look for this directly in the file system. But don't change anything. Otherwise, the conversion will no longer work. All very delicate. Actually, it's the best. Make a backup of the soul, the memory and the openclaw config every 10 minutes and try. Best way.

#

Not necessarily, there is a setting, my bot found it automatically. 'Cause I tell him, boy, stop making me poor... It worked.

thorny flint
still shell
#

can you run 2 seperate instances of clawdbot and keep private? my brother and i want to split a mac mini to test

strong sonnet
#

yes you can

vital wave
lean helm
#

So what I'm seeing Claude and Antigravity OAuth Methods are going to be blocked soon

past oxide
#

Question :
id you running an AWS instance , how can the agent emulate a chrome browser to do stuff on the web ? Or it just uses HTTP gets … ?

hidden latch
#

Hi all, I am not a coding expert but was so excited about all the reports regarding openclaw that I wanted to give it a try.

My aim is to do it on a Hetzner Server. Therefore, I followed the manual on the website.

I was able to install it and open the interface on my laptop, but it said pairing required. Unfortunately, I was not able to solve that issue.

Can anyone help me out?

strong sonnet
past oxide
sullen tree
#

My bot keeps saying the OpenClaw cron jobs are broken. I can’t get him to do anything proactively because the scheduled actions aren’t kicking in. Anyone else seeing this?

solar forge
#

my gateway keeps disconnecting randomly. i recently switched to codex 5.3

#

does anyone know how i can find the root cause?

grizzled mauve
static pasture
#

Hello/Morning! I have been struggling with a problem that prevents me from using openclaw - I hope some kind soul here can help:
OS: macOS Tahoe 26.2
Chip/Memory: M2 Max/64GB
Trying to run openclaw on docker. I cloned the repo and ran ./docker-setup.sh. I did the full setup, with local gateway/LAN (0.0.0.0:18789) using OpenAI free API key for testing. I skipped agents/skills setup, and the script finished. Then, I used docker compose to start the gateway. Here I run into the following problems:

  • the openclaw-cli container does not connect to the gateway with a token mismatch error. The token was auto generated by the setup script.
  • The control UI shows a keys-mismatch error if I maunally paste the key in ~/.openclaw/openclaw.json
  • The key in ~/.openclaw/openclaw.json matches the key in /home/node/.openclaw.json in the gateway container
    Any ideas how to troubleshoot this?
indigo robin
#

Hello everyone , new to Clawbot
How useful do you guys find clawbot .
What are daily uses of it .
Thanks

restive raven
#

Hi, I'm having an issue. I'm using the Gemini API for my AI project, but I'm running into problems because of the rate limits. Does anyone know of a good, free alternative?

cold drift
#

Does anyone here know how to circumvent the Claude Code OAuth to use the subscription for openclaw?

unkempt goblet
#

Does run Fedora Atomic + Podman any1?

harsh moss
#

do you guys use auto compaction or do you manually /compact?

#

I am having trouble getting auto compaction to function properly

rugged onyx
#

guys i cant use local ollama models on openclaw

harsh moss
#

me either, my tools come in as plain text, I think there is a chat template to address this but having trouble there too

digital minnow
#

hello when i start dowlanding i have a eror: node.exe : npm error code ENOENT
line:1 character:1

  • & "C:\Program Files\nodejs/node.exe" "C:\Program Files\nodejs/node_mo ...
  •   + CategoryInfo          : NotSpecified: (npm error code ENOENT:String) [], RemoteException
  + FullyQualifiedErrorId : NativeCommandError
mild lodge
# hidden latch Hi all, I am not a coding expert but was so excited about all the reports regard...

Whats your messaging channel? I have it up on telegram. After you install openclaw and gateway is running, (need to have a telegram bot already created, which is easy just go to botfather) configure messaging channel telegram, provide your telegram bot api key. Then message the bot on telegram. It will send you a pariing code, which then you need to use with openclaw pairing approve telegram <CODE>. Works the same for other channels (whatsapp, etc). Hopefully works out.

mild lodge
mild lodge
harsh moss
#

I keep trying "tool-capable" models but all they do is speak back to me and then SAY "{ tool: exec ..... }"

unkempt goblet
uneven forge
#

Hey guys so I am really unsure , but how can I make openclaw download voice notes from whatsapp using the whatsapp mac app?

digital minnow
digital minnow
keen canyon
#

Anyone have any luck with z.ai coding plan rate/concurrency limits? Seems like the best bet is to just rock with glm-4.7 and fall back on 4.7-flash or 4.6... just leaning on 4.7 seems to run into all sorts of issues.

harsh moss
#

am I supposed to use a different API mode when talking to an ollama endpoint or still just openai-completions?

unkempt goblet
coral herald
#

Does anyone know how to solve this problem:

Hey everyone, I'm stuck and really need some help.
I'm running OpenClaw on an AWS EC2 server and trying to connect it to Chrome on my local Windows machine using the relay extension. But no matter what I do, the extension just shows a red exclamation mark.
My setup:

Windows PC (local)
OpenClaw running on AWS EC2
Trying to get my Telegram bot to control Chrome so it can access my Shopify store

What I've tried:

Launched Chrome with --remote-debugging-port=9222
Killed all Chrome processes before relaunching
Allowed port 9222 in Windows Firewall (inbound rule, TCP)
Checked localhost:9222/json - connection refused
Tried both Program Files and Program Files (x86) paths

The problem:
Relay extension shows red ! and won't connect. Chrome debug port doesn't seem to be working either since localhost:9222 gives connection refused.
I've been at this for 2 days and I'm exhausted. Has anyone gotten this working with a similar setup (local Windows + remote server)?
Any help would be massively appreciated

?

mild lodge
# static pasture Hello/Morning! I have been struggling with a problem that prevents me from using...

This is very likely not a bad token, but might be the CLI is trying to authenticate against a different gateway identity. Quick checklist that usually fixes this:
Verify that all components use the SAME config
Host: ~/.openclaw/openclaw.json
Inside the gateway container: /home/node/.openclaw.json
cehck out volumes, too. Fix example: volumes: - ~/.openclaw:/home/node/.openclaw
Then restart cleanly: docker compose down -v && docker compose up
Check the gateway logs: docker logs <gateway-container> . If you see something like "generated new identity" then you definitely had an identity mismatch.
Fastest clean reset (if mismatch persists): rm -rf ~/.openclaw (use at your own risk though) && docker compose down -v && ./docker-setup.sh
Avoid mixing old identities with newly generated ones. OpenClaw tends to behave more predictably if you bootstrap in this order: gateway first → CLI second, instead of starting everything in parallel. Hope it works out.

loud compass
#

Hey everyone, yesterday I tried to optimize my openclaw but somehow spent 2.5 million token in...
Tried the /optimize extension and reseting my chat every two prompts but still it burned through my tokens.

Using claude sonnet

digital minnow
harsh moss
#

this might be the worst documentation in the world

static pasture
hardy venture
#

How do you install it on windows I entered the command in powershell

shrewd rover
#

I've found the reason of the failure of the cron jobs of the bot. THE HEARTBEAT.md FILE SHOULD NEVER BE EMPTY! Let the bot to write something in that file and the problem will just magically disappear with help of the latest release.

timid spindle
#

Hey! I just upgraded to Claude Opus 4.6 which supports 1M token context via the Anthropic API. I set agents.defaults.contextTokens: 1000000 in my config, but /status still shows 200k context. Is there a way to actually unlock the full 1M context window in OpenClaw? Or is there a hardcoded cap on the Anthropic provider side? Running v2026.2.6-3 on beta channel with direct Anthropic auth (not OpenRouter). Thanks!

harsh moss
#

its in the model list

#

you can override it though with mode merge

#

tui will show it after restarting tui and recieving a response

#

for the full example to merge (because it will not really merge, it overwrites so you need it all) go into the debug section of the web ui control panel and scroll down to the model list

timid spindle
#

oh, it worked actually LOL this is freaking nuts

shrewd rover